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 (td->io_ops && td->io_ops->get_zoned_model)
36 ret = td->io_ops->get_zoned_model(td, f, model);
38 ret = blkzoned_get_zoned_model(td, f, model);
40 td_verror(td, errno, "get zoned model failed");
41 log_err("%s: get zoned model failed (%d).\n",
49 * zbd_report_zones - Get zone information
50 * @td: FIO thread data.
51 * @f: FIO file for which to get zone information
52 * @offset: offset from which to report zones
53 * @zones: Array of struct zbd_zone
54 * @nr_zones: Size of @zones array
56 * Get zone information into @zones starting from the zone at offset @offset
57 * for the device specified by @f.
59 * Returns the number of zones reported upon success and a negative error code
60 * upon failure. If the zone report is empty, always assume an error (device
61 * problem) and return -EIO.
63 int zbd_report_zones(struct thread_data *td, struct fio_file *f,
64 uint64_t offset, struct zbd_zone *zones,
65 unsigned int nr_zones)
69 if (td->io_ops && td->io_ops->report_zones)
70 ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones);
72 ret = blkzoned_report_zones(td, f, offset, zones, nr_zones);
74 td_verror(td, errno, "report zones failed");
75 log_err("%s: report zones from sector %llu failed (%d).\n",
76 f->file_name, (unsigned long long)offset >> 9, errno);
77 } else if (ret == 0) {
78 td_verror(td, errno, "Empty zone report");
79 log_err("%s: report zones from sector %llu is empty.\n",
80 f->file_name, (unsigned long long)offset >> 9);
88 * zbd_reset_wp - reset the write pointer of a range of zones
89 * @td: FIO thread data.
90 * @f: FIO file for which to reset zones
91 * @offset: Starting offset of the first zone to reset
92 * @length: Length of the range of zones to reset
94 * Reset the write pointer of all zones in the range @offset...@offset+@length.
95 * Returns 0 upon success and a negative error code upon failure.
97 int zbd_reset_wp(struct thread_data *td, struct fio_file *f,
98 uint64_t offset, uint64_t length)
102 if (td->io_ops && td->io_ops->reset_wp)
103 ret = td->io_ops->reset_wp(td, f, offset, length);
105 ret = blkzoned_reset_wp(td, f, offset, length);
107 td_verror(td, errno, "resetting wp failed");
108 log_err("%s: resetting wp for %llu sectors at sector %llu failed (%d).\n",
109 f->file_name, (unsigned long long)length >> 9,
110 (unsigned long long)offset >> 9, errno);
117 * zbd_zone_idx - convert an offset into a zone number
119 * @offset: offset in bytes. If this offset is in the first zone_size bytes
120 * past the disk size then the index of the sentinel is returned.
122 static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset)
126 if (f->zbd_info->zone_size_log2 > 0)
127 zone_idx = offset >> f->zbd_info->zone_size_log2;
129 zone_idx = offset / f->zbd_info->zone_size;
131 return min(zone_idx, f->zbd_info->nr_zones);
135 * zbd_zone_end - Return zone end location
136 * @z: zone info pointer.
138 static inline uint64_t zbd_zone_end(const struct fio_zone_info *z)
144 * zbd_zone_capacity_end - Return zone capacity limit end location
145 * @z: zone info pointer.
147 static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z)
149 return z->start + z->capacity;
153 * zbd_zone_full - verify whether a minimum number of bytes remain in a zone
155 * @z: zone info pointer.
156 * @required: minimum number of bytes that must remain in a zone.
158 * The caller must hold z->mutex.
160 static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z,
163 assert((required & 511) == 0);
166 z->wp + required > zbd_zone_capacity_end(z);
169 static void zone_lock(struct thread_data *td, const struct fio_file *f,
170 struct fio_zone_info *z)
172 struct zoned_block_device_info *zbd = f->zbd_info;
173 uint32_t nz = z - zbd->zone_info;
175 /* A thread should never lock zones outside its working area. */
176 assert(f->min_zone <= nz && nz < f->max_zone);
181 * Lock the io_u target zone. The zone will be unlocked if io_u offset
182 * is changed or when io_u completes and zbd_put_io() executed.
183 * To avoid multiple jobs doing asynchronous I/Os from deadlocking each
184 * other waiting for zone locks when building an io_u batch, first
185 * only trylock the zone. If the zone is already locked by another job,
186 * process the currently queued I/Os so that I/O progress is made and
189 if (pthread_mutex_trylock(&z->mutex) != 0) {
190 if (!td_ioengine_flagged(td, FIO_SYNCIO))
192 pthread_mutex_lock(&z->mutex);
196 static inline void zone_unlock(struct fio_zone_info *z)
201 ret = pthread_mutex_unlock(&z->mutex);
205 static bool is_valid_offset(const struct fio_file *f, uint64_t offset)
207 return (uint64_t)(offset - f->file_offset) < f->io_size;
210 static inline struct fio_zone_info *get_zone(const struct fio_file *f,
211 unsigned int zone_nr)
213 return &f->zbd_info->zone_info[zone_nr];
216 /* Verify whether direct I/O is used for all host-managed zoned drives. */
217 static bool zbd_using_direct_io(void)
219 struct thread_data *td;
224 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
226 for_each_file(td, f, j) {
228 f->zbd_info->model == ZBD_HOST_MANAGED)
236 /* Whether or not the I/O range for f includes one or more sequential zones */
237 static bool zbd_is_seq_job(struct fio_file *f)
239 uint32_t zone_idx, zone_idx_b, zone_idx_e;
244 zone_idx_b = zbd_zone_idx(f, f->file_offset);
245 zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1);
246 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
247 if (get_zone(f, zone_idx)->has_wp)
254 * Verify whether offset and size parameters are aligned with zone boundaries.
256 static bool zbd_verify_sizes(void)
258 const struct fio_zone_info *z;
259 struct thread_data *td;
261 uint64_t new_offset, new_end;
266 for_each_file(td, f, j) {
269 if (f->file_offset >= f->real_file_size)
271 if (!zbd_is_seq_job(f))
274 if (!td->o.zone_size) {
275 td->o.zone_size = f->zbd_info->zone_size;
276 if (!td->o.zone_size) {
277 log_err("%s: invalid 0 zone size\n",
281 } else if (td->o.zone_size != f->zbd_info->zone_size) {
282 log_err("%s: job parameter zonesize %llu does not match disk zone size %llu.\n",
283 f->file_name, (unsigned long long) td->o.zone_size,
284 (unsigned long long) f->zbd_info->zone_size);
288 if (td->o.zone_skip % td->o.zone_size) {
289 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
290 f->file_name, (unsigned long long) td->o.zone_skip,
291 (unsigned long long) td->o.zone_size);
295 zone_idx = zbd_zone_idx(f, f->file_offset);
296 z = get_zone(f, zone_idx);
297 if ((f->file_offset != z->start) &&
298 (td->o.td_ddir != TD_DDIR_READ)) {
299 new_offset = zbd_zone_end(z);
300 if (new_offset >= f->file_offset + f->io_size) {
301 log_info("%s: io_size must be at least one zone\n",
305 log_info("%s: rounded up offset from %llu to %llu\n",
306 f->file_name, (unsigned long long) f->file_offset,
307 (unsigned long long) new_offset);
308 f->io_size -= (new_offset - f->file_offset);
309 f->file_offset = new_offset;
311 zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size);
312 z = get_zone(f, zone_idx);
314 if ((td->o.td_ddir != TD_DDIR_READ) &&
315 (f->file_offset + f->io_size != new_end)) {
316 if (new_end <= f->file_offset) {
317 log_info("%s: io_size must be at least one zone\n",
321 log_info("%s: rounded down io_size from %llu to %llu\n",
322 f->file_name, (unsigned long long) f->io_size,
323 (unsigned long long) new_end - f->file_offset);
324 f->io_size = new_end - f->file_offset;
332 static bool zbd_verify_bs(void)
334 struct thread_data *td;
340 for_each_file(td, f, j) {
343 zone_size = f->zbd_info->zone_size;
344 for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) {
345 if (td->o.verify != VERIFY_NONE &&
346 zone_size % td->o.bs[k] != 0) {
347 log_info("%s: block size %llu is not a divisor of the zone size %d\n",
348 f->file_name, td->o.bs[k],
358 static int ilog2(uint64_t i)
370 * Initialize f->zbd_info for devices that are not zoned block devices. This
371 * allows to execute a ZBD workload against a non-ZBD device.
373 static int init_zone_info(struct thread_data *td, struct fio_file *f)
376 struct fio_zone_info *p;
377 uint64_t zone_size = td->o.zone_size;
378 uint64_t zone_capacity = td->o.zone_capacity;
379 struct zoned_block_device_info *zbd_info = NULL;
382 if (zone_size == 0) {
383 log_err("%s: Specifying the zone size is mandatory for regular block devices with --zonemode=zbd\n\n",
388 if (zone_size < 512) {
389 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
394 if (zone_capacity == 0)
395 zone_capacity = zone_size;
397 if (zone_capacity > zone_size) {
398 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
399 f->file_name, (unsigned long long) td->o.zone_capacity,
400 (unsigned long long) td->o.zone_size);
404 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
405 zbd_info = scalloc(1, sizeof(*zbd_info) +
406 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
410 mutex_init_pshared(&zbd_info->mutex);
411 zbd_info->refcount = 1;
412 p = &zbd_info->zone_info[0];
413 for (i = 0; i < nr_zones; i++, p++) {
414 mutex_init_pshared_with_type(&p->mutex,
415 PTHREAD_MUTEX_RECURSIVE);
416 p->start = i * zone_size;
418 p->type = ZBD_ZONE_TYPE_SWR;
419 p->cond = ZBD_ZONE_COND_EMPTY;
420 p->capacity = zone_capacity;
424 p->start = nr_zones * zone_size;
426 f->zbd_info = zbd_info;
427 f->zbd_info->zone_size = zone_size;
428 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
429 ilog2(zone_size) : 0;
430 f->zbd_info->nr_zones = nr_zones;
435 * Maximum number of zones to report in one operation.
437 #define ZBD_REPORT_MAX_ZONES 8192U
440 * Parse the device zone report and store it in f->zbd_info. Must be called
441 * only for devices that are zoned, namely those with a model != ZBD_NONE.
443 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
446 struct zbd_zone *zones, *z;
447 struct fio_zone_info *p;
448 uint64_t zone_size, offset;
449 struct zoned_block_device_info *zbd_info = NULL;
450 int i, j, ret = -ENOMEM;
452 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
456 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
459 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
464 zone_size = zones[0].len;
465 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
467 if (td->o.zone_size == 0) {
468 td->o.zone_size = zone_size;
469 } else if (td->o.zone_size != zone_size) {
470 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %llu.\n",
471 f->file_name, (unsigned long long) td->o.zone_size,
472 (unsigned long long) zone_size);
477 dprint(FD_ZBD, "Device %s has %d zones of size %llu KB\n", f->file_name,
478 nr_zones, (unsigned long long) zone_size / 1024);
480 zbd_info = scalloc(1, sizeof(*zbd_info) +
481 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
484 mutex_init_pshared(&zbd_info->mutex);
485 zbd_info->refcount = 1;
486 p = &zbd_info->zone_info[0];
487 for (offset = 0, j = 0; j < nr_zones;) {
489 for (i = 0; i < nrz; i++, j++, z++, p++) {
490 mutex_init_pshared_with_type(&p->mutex,
491 PTHREAD_MUTEX_RECURSIVE);
493 p->capacity = z->capacity;
495 case ZBD_ZONE_COND_NOT_WP:
496 case ZBD_ZONE_COND_FULL:
497 p->wp = p->start + p->capacity;
500 assert(z->start <= z->wp);
501 assert(z->wp <= z->start + zone_size);
507 case ZBD_ZONE_TYPE_SWR:
516 if (j > 0 && p->start != p[-1].start + zone_size) {
517 log_info("%s: invalid zone data\n",
524 offset = z->start + z->len;
527 nrz = zbd_report_zones(td, f, offset, zones,
528 min((uint32_t)(nr_zones - j),
529 ZBD_REPORT_MAX_ZONES));
532 log_info("fio: report zones (offset %llu) failed for %s (%d).\n",
533 (unsigned long long)offset,
540 zbd_info->zone_info[nr_zones].start = offset;
542 f->zbd_info = zbd_info;
543 f->zbd_info->zone_size = zone_size;
544 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
545 ilog2(zone_size) : 0;
546 f->zbd_info->nr_zones = nr_zones;
557 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
559 * Returns 0 upon success and a negative error code upon failure.
561 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
563 enum zbd_zoned_model zbd_model;
566 assert(td->o.zone_mode == ZONE_MODE_ZBD);
568 ret = zbd_get_zoned_model(td, f, &zbd_model);
576 case ZBD_HOST_MANAGED:
577 ret = parse_zone_info(td, f);
580 ret = init_zone_info(td, f);
583 td_verror(td, EINVAL, "Unsupported zoned model");
584 log_err("Unsupported zoned model\n");
589 f->zbd_info->model = zbd_model;
590 f->zbd_info->max_open_zones = td->o.max_open_zones;
595 void zbd_free_zone_info(struct fio_file *f)
601 pthread_mutex_lock(&f->zbd_info->mutex);
602 refcount = --f->zbd_info->refcount;
603 pthread_mutex_unlock(&f->zbd_info->mutex);
605 assert((int32_t)refcount >= 0);
612 * Initialize f->zbd_info.
614 * Returns 0 upon success and a negative error code upon failure.
616 * Note: this function can only work correctly if it is called before the first
619 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
621 struct thread_data *td2;
625 for_each_td(td2, i) {
626 for_each_file(td2, f2, j) {
627 if (td2 == td && f2 == file)
630 strcmp(f2->file_name, file->file_name) != 0)
632 file->zbd_info = f2->zbd_info;
633 file->zbd_info->refcount++;
638 ret = zbd_create_zone_info(td, file);
640 td_verror(td, -ret, "zbd_create_zone_info() failed");
644 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
646 static int zbd_reset_zone(struct thread_data *td, struct fio_file *f,
647 struct fio_zone_info *z);
649 int zbd_setup_files(struct thread_data *td)
654 for_each_file(td, f, i) {
655 if (zbd_init_zone_info(td, f))
659 if (!zbd_using_direct_io()) {
660 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
664 if (!zbd_verify_sizes())
667 if (!zbd_verify_bs())
670 for_each_file(td, f, i) {
671 struct zoned_block_device_info *zbd = f->zbd_info;
672 struct fio_zone_info *z;
678 f->min_zone = zbd_zone_idx(f, f->file_offset);
679 f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size);
682 * When all zones in the I/O range are conventional, io_size
683 * can be smaller than zone size, making min_zone the same
684 * as max_zone. This is why the assert below needs to be made
687 if (zbd_is_seq_job(f))
688 assert(f->min_zone < f->max_zone);
690 zbd->max_open_zones = zbd->max_open_zones ?: ZBD_MAX_OPEN_ZONES;
692 if (td->o.max_open_zones > 0 &&
693 zbd->max_open_zones != td->o.max_open_zones) {
694 log_err("Different 'max_open_zones' values\n");
697 if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) {
698 log_err("'max_open_zones' value is limited by %u\n", ZBD_MAX_OPEN_ZONES);
702 for (zi = f->min_zone; zi < f->max_zone; zi++) {
703 z = &zbd->zone_info[zi];
704 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
705 z->cond != ZBD_ZONE_COND_EXP_OPEN)
707 if (zbd_open_zone(td, f, zi))
710 * If the number of open zones exceeds specified limits,
711 * reset all extra open zones.
713 if (zbd_reset_zone(td, f, z) < 0) {
714 log_err("Failed to reest zone %d\n", zi);
723 static inline unsigned int zbd_zone_nr(const struct fio_file *f,
724 struct fio_zone_info *zone)
726 return zone - f->zbd_info->zone_info;
730 * zbd_reset_zone - reset the write pointer of a single zone
731 * @td: FIO thread data.
732 * @f: FIO file associated with the disk for which to reset a write pointer.
735 * Returns 0 upon success and a negative error code upon failure.
737 * The caller must hold z->mutex.
739 static int zbd_reset_zone(struct thread_data *td, struct fio_file *f,
740 struct fio_zone_info *z)
742 uint64_t offset = z->start;
743 uint64_t length = (z+1)->start - offset;
744 uint64_t data_in_zone = z->wp - z->start;
750 assert(is_valid_offset(f, offset + length - 1));
752 dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", f->file_name,
754 switch (f->zbd_info->model) {
756 case ZBD_HOST_MANAGED:
757 ret = zbd_reset_wp(td, f, offset, length);
765 pthread_mutex_lock(&f->zbd_info->mutex);
766 f->zbd_info->sectors_with_data -= data_in_zone;
767 f->zbd_info->wp_sectors_with_data -= data_in_zone;
768 pthread_mutex_unlock(&f->zbd_info->mutex);
772 td->ts.nr_zone_resets++;
777 /* The caller must hold f->zbd_info->mutex */
778 static void zbd_close_zone(struct thread_data *td, const struct fio_file *f,
779 unsigned int zone_idx)
781 uint32_t open_zone_idx = 0;
783 for (; open_zone_idx < f->zbd_info->num_open_zones; open_zone_idx++) {
784 if (f->zbd_info->open_zones[open_zone_idx] == zone_idx)
787 if (open_zone_idx == f->zbd_info->num_open_zones)
790 dprint(FD_ZBD, "%s: closing zone %d\n", f->file_name, zone_idx);
791 memmove(f->zbd_info->open_zones + open_zone_idx,
792 f->zbd_info->open_zones + open_zone_idx + 1,
793 (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) *
794 sizeof(f->zbd_info->open_zones[0]));
795 f->zbd_info->num_open_zones--;
796 td->num_open_zones--;
797 get_zone(f, zone_idx)->open = 0;
801 * Reset a range of zones. Returns 0 upon success and 1 upon failure.
802 * @td: fio thread data.
803 * @f: fio file for which to reset zones
804 * @zb: first zone to reset.
805 * @ze: first zone not to reset.
806 * @all_zones: whether to reset all zones or only those zones for which the
807 * write pointer is not a multiple of td->o.min_bs[DDIR_WRITE].
809 static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
810 struct fio_zone_info *const zb,
811 struct fio_zone_info *const ze, bool all_zones)
813 struct fio_zone_info *z;
814 const uint32_t min_bs = td->o.min_bs[DDIR_WRITE];
820 dprint(FD_ZBD, "%s: examining zones %u .. %u\n", f->file_name,
821 zbd_zone_nr(f, zb), zbd_zone_nr(f, ze));
822 for (z = zb; z < ze; z++) {
823 uint32_t nz = zbd_zone_nr(f, z);
829 pthread_mutex_lock(&f->zbd_info->mutex);
830 zbd_close_zone(td, f, nz);
831 pthread_mutex_unlock(&f->zbd_info->mutex);
833 reset_wp = z->wp != z->start;
835 reset_wp = z->wp % min_bs != 0;
838 dprint(FD_ZBD, "%s: resetting zone %u\n",
839 f->file_name, zbd_zone_nr(f, z));
840 if (zbd_reset_zone(td, f, z) < 0)
850 * Reset zbd_info.write_cnt, the counter that counts down towards the next
853 static void _zbd_reset_write_cnt(const struct thread_data *td,
854 const struct fio_file *f)
856 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
858 f->zbd_info->write_cnt = td->o.zrf.u.f ?
859 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
862 static void zbd_reset_write_cnt(const struct thread_data *td,
863 const struct fio_file *f)
865 pthread_mutex_lock(&f->zbd_info->mutex);
866 _zbd_reset_write_cnt(td, f);
867 pthread_mutex_unlock(&f->zbd_info->mutex);
870 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
871 const struct fio_file *f)
873 uint32_t write_cnt = 0;
875 pthread_mutex_lock(&f->zbd_info->mutex);
876 assert(f->zbd_info->write_cnt);
877 if (f->zbd_info->write_cnt)
878 write_cnt = --f->zbd_info->write_cnt;
880 _zbd_reset_write_cnt(td, f);
881 pthread_mutex_unlock(&f->zbd_info->mutex);
883 return write_cnt == 0;
891 /* Calculate the number of sectors with data (swd) and perform action 'a' */
892 static uint64_t zbd_process_swd(struct thread_data *td,
893 const struct fio_file *f, enum swd_action a)
895 struct fio_zone_info *zb, *ze, *z;
899 zb = get_zone(f, f->min_zone);
900 ze = get_zone(f, f->max_zone);
901 for (z = zb; z < ze; z++) {
904 wp_swd += z->wp - z->start;
906 swd += z->wp - z->start;
908 pthread_mutex_lock(&f->zbd_info->mutex);
911 assert(f->zbd_info->sectors_with_data == swd);
912 assert(f->zbd_info->wp_sectors_with_data == wp_swd);
915 f->zbd_info->sectors_with_data = swd;
916 f->zbd_info->wp_sectors_with_data = wp_swd;
919 pthread_mutex_unlock(&f->zbd_info->mutex);
920 for (z = zb; z < ze; z++)
928 * The swd check is useful for debugging but takes too much time to leave
929 * it enabled all the time. Hence it is disabled by default.
931 static const bool enable_check_swd = false;
933 /* Check whether the values of zbd_info.*sectors_with_data are correct. */
934 static void zbd_check_swd(struct thread_data *td, const struct fio_file *f)
936 if (!enable_check_swd)
939 zbd_process_swd(td, f, CHECK_SWD);
942 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
944 struct fio_zone_info *zb, *ze;
947 if (!f->zbd_info || !td_write(td))
950 zb = get_zone(f, f->min_zone);
951 ze = get_zone(f, f->max_zone);
952 swd = zbd_process_swd(td, f, SET_SWD);
953 dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name,
956 * If data verification is enabled reset the affected zones before
957 * writing any data to avoid that a zone reset has to be issued while
958 * writing data, which causes data loss.
960 zbd_reset_zones(td, f, zb, ze, td->o.verify != VERIFY_NONE &&
961 td->runstate != TD_VERIFYING);
962 zbd_reset_write_cnt(td, f);
965 /* The caller must hold f->zbd_info->mutex. */
966 static bool is_zone_open(const struct thread_data *td, const struct fio_file *f,
967 unsigned int zone_idx)
969 struct zoned_block_device_info *zbdi = f->zbd_info;
972 assert(td->o.job_max_open_zones == 0 || td->num_open_zones <= td->o.job_max_open_zones);
973 assert(td->o.job_max_open_zones <= zbdi->max_open_zones);
974 assert(zbdi->num_open_zones <= zbdi->max_open_zones);
976 for (i = 0; i < zbdi->num_open_zones; i++)
977 if (zbdi->open_zones[i] == zone_idx)
984 * Open a ZBD zone if it was not yet open. Returns true if either the zone was
985 * already open or if opening a new zone is allowed. Returns false if the zone
986 * was not yet open and opening a new zone would cause the zone limit to be
989 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
992 const uint32_t min_bs = td->o.min_bs[DDIR_WRITE];
993 struct fio_zone_info *z = get_zone(f, zone_idx);
996 if (z->cond == ZBD_ZONE_COND_OFFLINE)
1000 * Skip full zones with data verification enabled because resetting a
1001 * zone causes data loss and hence causes verification to fail.
1003 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
1006 pthread_mutex_lock(&f->zbd_info->mutex);
1007 if (is_zone_open(td, f, zone_idx)) {
1009 * If the zone is already open and going to be full by writes
1010 * in-flight, handle it as a full zone instead of an open zone.
1012 if (z->wp >= zbd_zone_capacity_end(z))
1017 /* Zero means no limit */
1018 if (td->o.job_max_open_zones > 0 &&
1019 td->num_open_zones >= td->o.job_max_open_zones)
1021 if (f->zbd_info->num_open_zones >= f->zbd_info->max_open_zones)
1023 dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx);
1024 f->zbd_info->open_zones[f->zbd_info->num_open_zones++] = zone_idx;
1025 td->num_open_zones++;
1030 pthread_mutex_unlock(&f->zbd_info->mutex);
1034 /* Anything goes as long as it is not a constant. */
1035 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1036 const struct io_u *io_u)
1038 return io_u->offset * f->zbd_info->num_open_zones / f->real_file_size;
1042 * Modify the offset of an I/O unit that does not refer to an open zone such
1043 * that it refers to an open zone. Close an open zone and open a new zone if
1044 * necessary. The open zone is searched across sequential zones.
1045 * This algorithm can only work correctly if all write pointers are
1046 * a multiple of the fio block size. The caller must neither hold z->mutex
1047 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1049 static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
1052 const uint32_t min_bs = td->o.min_bs[io_u->ddir];
1053 struct fio_file *f = io_u->file;
1054 struct fio_zone_info *z;
1055 unsigned int open_zone_idx = -1;
1056 uint32_t zone_idx, new_zone_idx;
1058 bool wait_zone_close;
1060 assert(is_valid_offset(f, io_u->offset));
1062 if (td->o.max_open_zones || td->o.job_max_open_zones) {
1064 * This statement accesses f->zbd_info->open_zones[] on purpose
1067 zone_idx = f->zbd_info->open_zones[pick_random_zone_idx(f, io_u)];
1069 zone_idx = zbd_zone_idx(f, io_u->offset);
1071 if (zone_idx < f->min_zone)
1072 zone_idx = f->min_zone;
1073 else if (zone_idx >= f->max_zone)
1074 zone_idx = f->max_zone - 1;
1075 dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1076 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1079 * Since z->mutex is the outer lock and f->zbd_info->mutex the inner
1080 * lock it can happen that the state of the zone with index zone_idx
1081 * has changed after 'z' has been assigned and before f->zbd_info->mutex
1082 * has been obtained. Hence the loop.
1087 z = get_zone(f, zone_idx);
1089 zone_lock(td, f, z);
1090 pthread_mutex_lock(&f->zbd_info->mutex);
1092 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1093 td->o.max_open_zones == 0 && td->o.job_max_open_zones == 0)
1095 if (f->zbd_info->num_open_zones == 0) {
1096 dprint(FD_ZBD, "%s(%s): no zones are open\n",
1097 __func__, f->file_name);
1098 goto open_other_zone;
1103 * List of opened zones is per-device, shared across all threads.
1104 * Start with quasi-random candidate zone.
1105 * Ignore zones which don't belong to thread's offset/size area.
1107 open_zone_idx = pick_random_zone_idx(f, io_u);
1108 assert(!open_zone_idx ||
1109 open_zone_idx < f->zbd_info->num_open_zones);
1110 tmp_idx = open_zone_idx;
1111 for (i = 0; i < f->zbd_info->num_open_zones; i++) {
1114 if (tmp_idx >= f->zbd_info->num_open_zones)
1116 tmpz = f->zbd_info->open_zones[tmp_idx];
1117 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1118 open_zone_idx = tmp_idx;
1119 goto found_candidate_zone;
1125 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1126 __func__, f->file_name);
1127 pthread_mutex_unlock(&f->zbd_info->mutex);
1132 found_candidate_zone:
1133 new_zone_idx = f->zbd_info->open_zones[open_zone_idx];
1134 if (new_zone_idx == zone_idx)
1136 zone_idx = new_zone_idx;
1137 pthread_mutex_unlock(&f->zbd_info->mutex);
1142 /* Both z->mutex and f->zbd_info->mutex are held. */
1145 if (z->wp + min_bs <= zbd_zone_capacity_end(z)) {
1146 pthread_mutex_unlock(&f->zbd_info->mutex);
1151 /* Check if number of open zones reaches one of limits. */
1153 f->zbd_info->num_open_zones == f->max_zone - f->min_zone ||
1154 (td->o.max_open_zones &&
1155 f->zbd_info->num_open_zones == td->o.max_open_zones) ||
1156 (td->o.job_max_open_zones &&
1157 td->num_open_zones == td->o.job_max_open_zones);
1159 pthread_mutex_unlock(&f->zbd_info->mutex);
1161 /* Only z->mutex is held. */
1164 * When number of open zones reaches to one of limits, wait for
1165 * zone close before opening a new zone.
1167 if (wait_zone_close) {
1168 dprint(FD_ZBD, "%s(%s): quiesce to allow open zones to close\n",
1169 __func__, f->file_name);
1173 /* Zone 'z' is full, so try to open a new zone. */
1174 for (i = f->io_size / f->zbd_info->zone_size; i > 0; i--) {
1179 if (!is_valid_offset(f, z->start)) {
1181 zone_idx = f->min_zone;
1182 z = get_zone(f, zone_idx);
1184 assert(is_valid_offset(f, z->start));
1187 zone_lock(td, f, z);
1190 if (zbd_open_zone(td, f, zone_idx))
1194 /* Only z->mutex is held. */
1196 /* Check whether the write fits in any of the already opened zones. */
1197 pthread_mutex_lock(&f->zbd_info->mutex);
1198 for (i = 0; i < f->zbd_info->num_open_zones; i++) {
1199 zone_idx = f->zbd_info->open_zones[i];
1200 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1202 pthread_mutex_unlock(&f->zbd_info->mutex);
1205 z = get_zone(f, zone_idx);
1207 zone_lock(td, f, z);
1208 if (z->wp + min_bs <= zbd_zone_capacity_end(z))
1210 pthread_mutex_lock(&f->zbd_info->mutex);
1212 pthread_mutex_unlock(&f->zbd_info->mutex);
1214 dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__,
1219 dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name,
1221 io_u->offset = z->start;
1223 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1227 /* The caller must hold z->mutex. */
1228 static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td,
1230 struct fio_zone_info *z)
1232 const struct fio_file *f = io_u->file;
1233 const uint32_t min_bs = td->o.min_bs[DDIR_WRITE];
1235 if (!zbd_open_zone(td, f, zbd_zone_nr(f, z))) {
1237 z = zbd_convert_to_open_zone(td, io_u);
1241 if (z->verify_block * min_bs >= z->capacity) {
1242 log_err("%s: %d * %d >= %llu\n", f->file_name, z->verify_block,
1243 min_bs, (unsigned long long)z->capacity);
1245 * If the assertion below fails during a test run, adding
1246 * "--experimental_verify=1" to the command line may help.
1250 io_u->offset = z->start + z->verify_block * min_bs;
1251 if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1252 log_err("%s: %llu + %llu >= %llu\n", f->file_name, io_u->offset,
1253 io_u->buflen, (unsigned long long) zbd_zone_capacity_end(z));
1256 z->verify_block += io_u->buflen / min_bs;
1262 * Find another zone for which @io_u fits in the readable data in the zone.
1263 * Search in zones @zb + 1 .. @zl. For random workload, also search in zones
1266 * Either returns NULL or returns a zone pointer. When the zone has write
1267 * pointer, hold the mutex for the zone.
1269 static struct fio_zone_info *
1270 zbd_find_zone(struct thread_data *td, struct io_u *io_u,
1271 struct fio_zone_info *zb, struct fio_zone_info *zl)
1273 const uint32_t min_bs = td->o.min_bs[io_u->ddir];
1274 struct fio_file *f = io_u->file;
1275 struct fio_zone_info *z1, *z2;
1276 const struct fio_zone_info *const zf = get_zone(f, f->min_zone);
1279 * Skip to the next non-empty zone in case of sequential I/O and to
1280 * the nearest non-empty zone in case of random I/O.
1282 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1283 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1285 zone_lock(td, f, z1);
1286 if (z1->start + min_bs <= z1->wp)
1290 } else if (!td_random(td)) {
1293 if (td_random(td) && z2 >= zf &&
1294 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1296 zone_lock(td, f, z2);
1297 if (z2->start + min_bs <= z2->wp)
1303 dprint(FD_ZBD, "%s: adjusting random read offset failed\n",
1309 * zbd_end_zone_io - update zone status at command completion
1311 * @z: zone info pointer
1313 * If the write command made the zone full, close it.
1315 * The caller must hold z->mutex.
1317 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1318 struct fio_zone_info *z)
1320 const struct fio_file *f = io_u->file;
1322 if (io_u->ddir == DDIR_WRITE &&
1323 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1324 pthread_mutex_lock(&f->zbd_info->mutex);
1325 zbd_close_zone(td, f, zbd_zone_nr(f, z));
1326 pthread_mutex_unlock(&f->zbd_info->mutex);
1331 * zbd_queue_io - update the write pointer of a sequential zone
1333 * @success: Whether or not the I/O unit has been queued successfully
1334 * @q: queueing status (busy, completed or queued).
1336 * For write and trim operations, update the write pointer of the I/O unit
1339 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1342 const struct fio_file *f = io_u->file;
1343 struct zoned_block_device_info *zbd_info = f->zbd_info;
1344 struct fio_zone_info *z;
1351 zone_idx = zbd_zone_idx(f, io_u->offset);
1352 assert(zone_idx < zbd_info->nr_zones);
1353 z = get_zone(f, zone_idx);
1361 "%s: queued I/O (%lld, %llu) for zone %u\n",
1362 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1364 switch (io_u->ddir) {
1366 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1367 zbd_zone_capacity_end(z));
1368 pthread_mutex_lock(&zbd_info->mutex);
1370 * z->wp > zone_end means that one or more I/O errors
1373 if (z->wp <= zone_end) {
1374 zbd_info->sectors_with_data += zone_end - z->wp;
1375 zbd_info->wp_sectors_with_data += zone_end - z->wp;
1377 pthread_mutex_unlock(&zbd_info->mutex);
1381 assert(z->wp == z->start);
1387 if (q == FIO_Q_COMPLETED && !io_u->error)
1388 zbd_end_zone_io(td, io_u, z);
1391 if (!success || q != FIO_Q_QUEUED) {
1392 /* BUSY or COMPLETED: unlock the zone */
1394 io_u->zbd_put_io = NULL;
1399 * zbd_put_io - Unlock an I/O unit target zone lock
1402 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1404 const struct fio_file *f = io_u->file;
1405 struct zoned_block_device_info *zbd_info = f->zbd_info;
1406 struct fio_zone_info *z;
1412 zone_idx = zbd_zone_idx(f, io_u->offset);
1413 assert(zone_idx < zbd_info->nr_zones);
1414 z = get_zone(f, zone_idx);
1419 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1420 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1422 zbd_end_zone_io(td, io_u, z);
1425 zbd_check_swd(td, f);
1429 * Windows and MacOS do not define this.
1432 #define EREMOTEIO 121 /* POSIX value */
1435 bool zbd_unaligned_write(int error_code)
1437 switch (error_code) {
1446 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1447 * @td: FIO thread data.
1448 * @io_u: FIO I/O unit.
1450 * For sequential workloads, change the file offset to skip zoneskip bytes when
1451 * no more IO can be performed in the current zone.
1452 * - For read workloads, zoneskip is applied when the io has reached the end of
1453 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1454 * - For write workloads, zoneskip is applied when the zone is full.
1455 * This applies only to read and write operations.
1457 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1459 struct fio_file *f = io_u->file;
1460 enum fio_ddir ddir = io_u->ddir;
1461 struct fio_zone_info *z;
1464 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1465 assert(td->o.zone_size);
1467 zone_idx = zbd_zone_idx(f, f->last_pos[ddir]);
1468 z = get_zone(f, zone_idx);
1471 * When the zone capacity is smaller than the zone size and the I/O is
1472 * sequential write, skip to zone end if the latest position is at the
1473 * zone capacity limit.
1475 if (z->capacity < f->zbd_info->zone_size && !td_random(td) &&
1476 ddir == DDIR_WRITE &&
1477 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1479 "%s: Jump from zone capacity limit to zone end:"
1480 " (%llu -> %llu) for zone %u (%llu)\n",
1481 f->file_name, (unsigned long long) f->last_pos[ddir],
1482 (unsigned long long) zbd_zone_end(z), zone_idx,
1483 (unsigned long long) z->capacity);
1484 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1485 f->last_pos[ddir] = zbd_zone_end(z);
1489 * zone_skip is valid only for sequential workloads.
1491 if (td_random(td) || !td->o.zone_skip)
1495 * It is time to switch to a new zone if:
1496 * - zone_bytes == zone_size bytes have already been accessed
1497 * - The last position reached the end of the current zone.
1498 * - For reads with td->o.read_beyond_wp == false, the last position
1499 * reached the zone write pointer.
1501 if (td->zone_bytes >= td->o.zone_size ||
1502 f->last_pos[ddir] >= zbd_zone_end(z) ||
1503 (ddir == DDIR_READ &&
1504 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1509 f->file_offset += td->o.zone_size + td->o.zone_skip;
1512 * Wrap from the beginning, if we exceed the file size
1514 if (f->file_offset >= f->real_file_size)
1515 f->file_offset = get_start_offset(td, f);
1517 f->last_pos[ddir] = f->file_offset;
1518 td->io_skip_bytes += td->o.zone_skip;
1523 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1525 * @td: FIO thread data.
1526 * @io_u: FIO I/O unit.
1527 * @ddir: I/O direction before adjustment.
1529 * Return adjusted I/O direction.
1531 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1535 * In case read direction is chosen for the first random I/O, fio with
1536 * zonemode=zbd stops because no data can be read from zoned block
1537 * devices with all empty zones. Overwrite the first I/O direction as
1538 * write to make sure data to read exists.
1540 if (ddir != DDIR_READ || !td_rw(td))
1543 if (io_u->file->zbd_info->sectors_with_data ||
1544 td->o.read_beyond_wp)
1551 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1552 * @td: FIO thread data.
1553 * @io_u: FIO I/O unit.
1555 * Locking strategy: returns with z->mutex locked if and only if z refers
1556 * to a sequential zone and if io_u_accept is returned. z is the zone that
1557 * corresponds to io_u->offset at the end of this function.
1559 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1561 struct fio_file *f = io_u->file;
1562 uint32_t zone_idx_b;
1563 struct fio_zone_info *zb, *zl, *orig_zb;
1564 uint32_t orig_len = io_u->buflen;
1565 uint32_t min_bs = td->o.min_bs[io_u->ddir];
1573 assert(is_valid_offset(f, io_u->offset));
1574 assert(io_u->buflen);
1575 zone_idx_b = zbd_zone_idx(f, io_u->offset);
1576 zb = get_zone(f, zone_idx_b);
1580 /* Accept non-write I/Os for conventional zones. */
1581 if (io_u->ddir != DDIR_WRITE)
1584 * Make sure that writes to conventional zones
1585 * don't cross over to any sequential zones.
1587 if (!(zb + 1)->has_wp ||
1588 io_u->offset + io_u->buflen <= (zb + 1)->start)
1591 if (io_u->offset + min_bs > (zb + 1)->start) {
1593 "%s: off=%llu + min_bs=%u > next zone %llu\n",
1594 f->file_name, io_u->offset,
1595 min_bs, (unsigned long long) (zb + 1)->start);
1596 io_u->offset = zb->start + (zb + 1)->start - io_u->offset;
1597 new_len = min(io_u->buflen, (zb + 1)->start - io_u->offset);
1599 new_len = (zb + 1)->start - io_u->offset;
1601 io_u->buflen = new_len / min_bs * min_bs;
1606 * Accept the I/O offset for reads if reading beyond the write pointer
1609 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1610 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1613 zbd_check_swd(td, f);
1615 zone_lock(td, f, zb);
1617 switch (io_u->ddir) {
1619 if (td->runstate == TD_VERIFYING && td_write(td)) {
1620 zb = zbd_replay_write_order(td, io_u, zb);
1624 * Check that there is enough written data in the zone to do an
1625 * I/O of at least min_bs B. If there isn't, find a new zone for
1628 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1629 zb->wp - zb->start : 0;
1630 if (range < min_bs ||
1631 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1633 zl = get_zone(f, f->max_zone);
1634 zb = zbd_find_zone(td, io_u, zb, zl);
1637 "%s: zbd_find_zone(%lld, %llu) failed\n",
1638 f->file_name, io_u->offset,
1643 * zbd_find_zone() returned a zone with a range of at
1646 range = zb->wp - zb->start;
1647 assert(range >= min_bs);
1650 io_u->offset = zb->start;
1653 * Make sure the I/O is within the zone valid data range while
1654 * maximizing the I/O size and preserving randomness.
1656 if (range <= io_u->buflen)
1657 io_u->offset = zb->start;
1658 else if (td_random(td))
1659 io_u->offset = zb->start +
1660 ((io_u->offset - orig_zb->start) %
1661 (range - io_u->buflen)) / min_bs * min_bs;
1663 * When zbd_find_zone() returns a conventional zone,
1664 * we can simply accept the new i/o offset here.
1669 * Make sure the I/O does not cross over the zone wp position.
1671 new_len = min((unsigned long long)io_u->buflen,
1672 (unsigned long long)(zb->wp - io_u->offset));
1673 new_len = new_len / min_bs * min_bs;
1674 if (new_len < io_u->buflen) {
1675 io_u->buflen = new_len;
1676 dprint(FD_IO, "Changed length from %u into %llu\n",
1677 orig_len, io_u->buflen);
1679 assert(zb->start <= io_u->offset);
1680 assert(io_u->offset + io_u->buflen <= zb->wp);
1683 if (io_u->buflen > f->zbd_info->zone_size) {
1684 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
1686 "%s: I/O buflen %llu exceeds zone size %llu\n",
1687 f->file_name, io_u->buflen,
1688 (unsigned long long) f->zbd_info->zone_size);
1691 if (!zbd_open_zone(td, f, zone_idx_b)) {
1693 zb = zbd_convert_to_open_zone(td, io_u);
1695 dprint(FD_IO, "%s: can't convert to open zone",
1699 zone_idx_b = zbd_zone_nr(f, zb);
1701 /* Check whether the zone reset threshold has been exceeded */
1702 if (td->o.zrf.u.f) {
1703 if (f->zbd_info->wp_sectors_with_data >=
1704 f->io_size * td->o.zrt.u.f &&
1705 zbd_dec_and_reset_write_cnt(td, f)) {
1709 /* Reset the zone pointer if necessary */
1710 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
1711 assert(td->o.verify == VERIFY_NONE);
1713 * Since previous write requests may have been submitted
1714 * asynchronously and since we will submit the zone
1715 * reset synchronously, wait until previously submitted
1716 * write requests have completed before issuing a
1721 if (zbd_reset_zone(td, f, zb) < 0)
1724 if (zb->capacity < min_bs) {
1725 td_verror(td, EINVAL, "ZCAP is less min_bs");
1726 log_err("zone capacity %llu smaller than minimum block size %d\n",
1727 (unsigned long long)zb->capacity,
1732 /* Make writes occur at the write pointer */
1733 assert(!zbd_zone_full(f, zb, min_bs));
1734 io_u->offset = zb->wp;
1735 if (!is_valid_offset(f, io_u->offset)) {
1736 td_verror(td, EINVAL, "invalid WP value");
1737 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
1738 f->file_name, io_u->offset);
1742 * Make sure that the buflen is a multiple of the minimal
1743 * block size. Give up if shrinking would make the request too
1746 new_len = min((unsigned long long)io_u->buflen,
1747 zbd_zone_capacity_end(zb) - io_u->offset);
1748 new_len = new_len / min_bs * min_bs;
1749 if (new_len == io_u->buflen)
1751 if (new_len >= min_bs) {
1752 io_u->buflen = new_len;
1753 dprint(FD_IO, "Changed length from %u into %llu\n",
1754 orig_len, io_u->buflen);
1757 td_verror(td, EIO, "zone remainder too small");
1758 log_err("zone remainder %lld smaller than min block size %d\n",
1759 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
1765 case DDIR_SYNC_FILE_RANGE:
1776 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
1777 assert(!io_u->zbd_queue_io);
1778 assert(!io_u->zbd_put_io);
1779 io_u->zbd_queue_io = zbd_queue_io;
1780 io_u->zbd_put_io = zbd_put_io;
1782 * Since we return with the zone lock still held,
1783 * add an annotation to let Coverity know that it
1786 /* coverity[missing_unlock] */
1790 if (zb && zb->has_wp)
1795 /* Return a string with ZBD statistics */
1796 char *zbd_write_status(const struct thread_stat *ts)
1800 if (asprintf(&res, "; %llu zone resets", (unsigned long long) ts->nr_zone_resets) < 0)