2 * Copyright (C) 2018 Western Digital Corporation or its affiliates.
4 * This file is released under the GPL.
19 #include "oslib/asprintf.h"
25 static bool is_valid_offset(const struct fio_file *f, uint64_t offset)
27 return (uint64_t)(offset - f->file_offset) < f->io_size;
30 static inline unsigned int zbd_zone_idx(const struct fio_file *f,
31 struct fio_zone_info *zone)
33 return zone - f->zbd_info->zone_info;
37 * zbd_offset_to_zone_idx - convert an offset into a zone number
39 * @offset: offset in bytes. If this offset is in the first zone_size bytes
40 * past the disk size then the index of the sentinel is returned.
42 static unsigned int zbd_offset_to_zone_idx(const struct fio_file *f,
47 if (f->zbd_info->zone_size_log2 > 0)
48 zone_idx = offset >> f->zbd_info->zone_size_log2;
50 zone_idx = offset / f->zbd_info->zone_size;
52 return min(zone_idx, f->zbd_info->nr_zones);
56 * zbd_zone_end - Return zone end location
57 * @z: zone info pointer.
59 static inline uint64_t zbd_zone_end(const struct fio_zone_info *z)
65 * zbd_zone_capacity_end - Return zone capacity limit end location
66 * @z: zone info pointer.
68 static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z)
70 return z->start + z->capacity;
74 * zbd_zone_remainder - Return the number of bytes that are still available for
75 * writing before the zone gets full
76 * @z: zone info pointer.
78 static inline uint64_t zbd_zone_remainder(struct fio_zone_info *z)
80 if (z->wp >= zbd_zone_capacity_end(z))
83 return zbd_zone_capacity_end(z) - z->wp;
87 * zbd_zone_full - verify whether a minimum number of bytes remain in a zone
89 * @z: zone info pointer.
90 * @required: minimum number of bytes that must remain in a zone.
92 * The caller must hold z->mutex.
94 static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z,
97 assert((required & 511) == 0);
99 return z->has_wp && required > zbd_zone_remainder(z);
102 static void zone_lock(struct thread_data *td, const struct fio_file *f,
103 struct fio_zone_info *z)
105 struct zoned_block_device_info *zbd = f->zbd_info;
106 uint32_t nz = z - zbd->zone_info;
108 /* A thread should never lock zones outside its working area. */
109 assert(f->min_zone <= nz && nz < f->max_zone);
114 * Lock the io_u target zone. The zone will be unlocked if io_u offset
115 * is changed or when io_u completes and zbd_put_io() executed.
116 * To avoid multiple jobs doing asynchronous I/Os from deadlocking each
117 * other waiting for zone locks when building an io_u batch, first
118 * only trylock the zone. If the zone is already locked by another job,
119 * process the currently queued I/Os so that I/O progress is made and
122 if (pthread_mutex_trylock(&z->mutex) != 0) {
123 if (!td_ioengine_flagged(td, FIO_SYNCIO))
125 pthread_mutex_lock(&z->mutex);
129 static inline void zone_unlock(struct fio_zone_info *z)
134 ret = pthread_mutex_unlock(&z->mutex);
138 static inline struct fio_zone_info *zbd_get_zone(const struct fio_file *f,
139 unsigned int zone_idx)
141 return &f->zbd_info->zone_info[zone_idx];
144 static inline struct fio_zone_info *
145 zbd_offset_to_zone(const struct fio_file *f, uint64_t offset)
147 return zbd_get_zone(f, zbd_offset_to_zone_idx(f, offset));
150 static bool accounting_vdb(struct thread_data *td, const struct fio_file *f)
152 return td->o.zrt.u.f && td_write(td);
156 * zbd_get_zoned_model - Get a device zoned model
157 * @td: FIO thread data
158 * @f: FIO file for which to get model information
160 static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f,
161 enum zbd_zoned_model *model)
165 if (f->filetype == FIO_TYPE_PIPE) {
166 log_err("zonemode=zbd does not support pipes\n");
170 /* If regular file, always emulate zones inside the file. */
171 if (f->filetype == FIO_TYPE_FILE) {
176 if (td->io_ops && td->io_ops->get_zoned_model)
177 ret = td->io_ops->get_zoned_model(td, f, model);
179 ret = blkzoned_get_zoned_model(td, f, model);
181 td_verror(td, errno, "get zoned model failed");
182 log_err("%s: get zoned model failed (%d).\n",
183 f->file_name, errno);
190 * zbd_report_zones - Get zone information
191 * @td: FIO thread data.
192 * @f: FIO file for which to get zone information
193 * @offset: offset from which to report zones
194 * @zones: Array of struct zbd_zone
195 * @nr_zones: Size of @zones array
197 * Get zone information into @zones starting from the zone at offset @offset
198 * for the device specified by @f.
200 * Returns the number of zones reported upon success and a negative error code
201 * upon failure. If the zone report is empty, always assume an error (device
202 * problem) and return -EIO.
204 static int zbd_report_zones(struct thread_data *td, struct fio_file *f,
205 uint64_t offset, struct zbd_zone *zones,
206 unsigned int nr_zones)
210 if (td->io_ops && td->io_ops->report_zones)
211 ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones);
213 ret = blkzoned_report_zones(td, f, offset, zones, nr_zones);
215 td_verror(td, errno, "report zones failed");
216 log_err("%s: report zones from sector %"PRIu64" failed (%d).\n",
217 f->file_name, offset >> 9, errno);
218 } else if (ret == 0) {
219 td_verror(td, errno, "Empty zone report");
220 log_err("%s: report zones from sector %"PRIu64" is empty.\n",
221 f->file_name, offset >> 9);
229 * zbd_reset_wp - reset the write pointer of a range of zones
230 * @td: FIO thread data.
231 * @f: FIO file for which to reset zones
232 * @offset: Starting offset of the first zone to reset
233 * @length: Length of the range of zones to reset
235 * Reset the write pointer of all zones in the range @offset...@offset+@length.
236 * Returns 0 upon success and a negative error code upon failure.
238 static int zbd_reset_wp(struct thread_data *td, struct fio_file *f,
239 uint64_t offset, uint64_t length)
243 if (td->io_ops && td->io_ops->reset_wp)
244 ret = td->io_ops->reset_wp(td, f, offset, length);
246 ret = blkzoned_reset_wp(td, f, offset, length);
248 td_verror(td, errno, "resetting wp failed");
249 log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n",
250 f->file_name, length >> 9, offset >> 9, errno);
257 * zbd_reset_zone - reset the write pointer of a single zone
258 * @td: FIO thread data.
259 * @f: FIO file associated with the disk for which to reset a write pointer.
262 * Returns 0 upon success and a negative error code upon failure.
264 * The caller must hold z->mutex.
266 static int zbd_reset_zone(struct thread_data *td, struct fio_file *f,
267 struct fio_zone_info *z)
269 uint64_t offset = z->start;
270 uint64_t length = (z+1)->start - offset;
271 uint64_t data_in_zone = z->wp - z->start;
277 assert(is_valid_offset(f, offset + length - 1));
279 dprint(FD_ZBD, "%s: resetting wp of zone %u.\n",
280 f->file_name, zbd_zone_idx(f, z));
282 switch (f->zbd_info->model) {
284 case ZBD_HOST_MANAGED:
285 ret = zbd_reset_wp(td, f, offset, length);
293 if (accounting_vdb(td, f)) {
294 pthread_mutex_lock(&f->zbd_info->mutex);
295 f->zbd_info->wp_valid_data_bytes -= data_in_zone;
296 pthread_mutex_unlock(&f->zbd_info->mutex);
301 td->ts.nr_zone_resets++;
307 * zbd_close_zone - Remove a zone from the open zones array.
308 * @td: FIO thread data.
309 * @f: FIO file associated with the disk for which to reset a write pointer.
310 * @zone_idx: Index of the zone to remove.
312 * The caller must hold f->zbd_info->mutex.
314 static void zbd_close_zone(struct thread_data *td, const struct fio_file *f,
315 struct fio_zone_info *z)
322 for (ozi = 0; ozi < f->zbd_info->num_open_zones; ozi++) {
323 if (zbd_get_zone(f, f->zbd_info->open_zones[ozi]) == z)
326 if (ozi == f->zbd_info->num_open_zones)
329 dprint(FD_ZBD, "%s: closing zone %u\n",
330 f->file_name, zbd_zone_idx(f, z));
332 memmove(f->zbd_info->open_zones + ozi,
333 f->zbd_info->open_zones + ozi + 1,
334 (ZBD_MAX_OPEN_ZONES - (ozi + 1)) *
335 sizeof(f->zbd_info->open_zones[0]));
337 f->zbd_info->num_open_zones--;
338 td->num_open_zones--;
343 * zbd_finish_zone - finish the specified zone
344 * @td: FIO thread data.
345 * @f: FIO file for which to finish a zone
346 * @z: Zone to finish.
348 * Finish the zone at @offset with open or close status.
350 static int zbd_finish_zone(struct thread_data *td, struct fio_file *f,
351 struct fio_zone_info *z)
353 uint64_t offset = z->start;
354 uint64_t length = f->zbd_info->zone_size;
357 switch (f->zbd_info->model) {
359 case ZBD_HOST_MANAGED:
360 if (td->io_ops && td->io_ops->finish_zone)
361 ret = td->io_ops->finish_zone(td, f, offset, length);
363 ret = blkzoned_finish_zone(td, f, offset, length);
370 td_verror(td, errno, "finish zone failed");
371 log_err("%s: finish zone at sector %"PRIu64" failed (%d).\n",
372 f->file_name, offset >> 9, errno);
374 z->wp = (z+1)->start;
381 * zbd_reset_zones - Reset a range of zones.
382 * @td: fio thread data.
383 * @f: fio file for which to reset zones
384 * @zb: first zone to reset.
385 * @ze: first zone not to reset.
387 * Returns 0 upon success and 1 upon failure.
389 static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
390 struct fio_zone_info *const zb,
391 struct fio_zone_info *const ze)
393 struct fio_zone_info *z;
394 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
399 dprint(FD_ZBD, "%s: examining zones %u .. %u\n",
400 f->file_name, zbd_zone_idx(f, zb), zbd_zone_idx(f, ze));
402 for (z = zb; z < ze; z++) {
407 pthread_mutex_lock(&f->zbd_info->mutex);
408 zbd_close_zone(td, f, z);
409 pthread_mutex_unlock(&f->zbd_info->mutex);
411 if (z->wp != z->start) {
412 dprint(FD_ZBD, "%s: resetting zone %u\n",
413 f->file_name, zbd_zone_idx(f, z));
414 if (zbd_reset_zone(td, f, z) < 0)
425 * zbd_get_max_open_zones - Get the maximum number of open zones
426 * @td: FIO thread data
427 * @f: FIO file for which to get max open zones
428 * @max_open_zones: Upon success, result will be stored here.
430 * A @max_open_zones value set to zero means no limit.
432 * Returns 0 upon success and a negative error code upon failure.
434 static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f,
435 unsigned int *max_open_zones)
439 if (td->io_ops && td->io_ops->get_max_open_zones)
440 ret = td->io_ops->get_max_open_zones(td, f, max_open_zones);
442 ret = blkzoned_get_max_open_zones(td, f, max_open_zones);
444 td_verror(td, errno, "get max open zones failed");
445 log_err("%s: get max open zones failed (%d).\n",
446 f->file_name, errno);
453 * zbd_open_zone - Add a zone to the array of open zones.
454 * @td: fio thread data.
455 * @f: fio file that has the open zones to add.
456 * @zone_idx: Index of the zone to add.
458 * Open a ZBD zone if it is not already open. Returns true if either the zone
459 * was already open or if the zone was successfully added to the array of open
460 * zones without exceeding the maximum number of open zones. Returns false if
461 * the zone was not already open and opening the zone would cause the zone limit
464 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
465 struct fio_zone_info *z)
467 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
468 struct zoned_block_device_info *zbdi = f->zbd_info;
469 uint32_t zone_idx = zbd_zone_idx(f, z);
472 if (z->cond == ZBD_ZONE_COND_OFFLINE)
476 * Skip full zones with data verification enabled because resetting a
477 * zone causes data loss and hence causes verification to fail.
479 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
483 * zbdi->max_open_zones == 0 means that there is no limit on the maximum
484 * number of open zones. In this case, do no track open zones in
485 * zbdi->open_zones array.
487 if (!zbdi->max_open_zones)
490 pthread_mutex_lock(&zbdi->mutex);
494 * If the zone is going to be completely filled by writes
495 * already in-flight, handle it as a full zone instead of an
498 if (!zbd_zone_remainder(z))
504 /* Zero means no limit */
505 if (td->o.job_max_open_zones > 0 &&
506 td->num_open_zones >= td->o.job_max_open_zones)
508 if (zbdi->num_open_zones >= zbdi->max_open_zones)
511 dprint(FD_ZBD, "%s: opening zone %u\n",
512 f->file_name, zone_idx);
514 zbdi->open_zones[zbdi->num_open_zones++] = zone_idx;
515 td->num_open_zones++;
520 pthread_mutex_unlock(&zbdi->mutex);
524 /* Verify whether direct I/O is used for all host-managed zoned block drives. */
525 static bool zbd_using_direct_io(void)
527 struct thread_data *td;
532 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
534 for_each_file(td, f, j) {
535 if (f->zbd_info && f->filetype == FIO_TYPE_BLOCK &&
536 f->zbd_info->model == ZBD_HOST_MANAGED)
544 /* Whether or not the I/O range for f includes one or more sequential zones */
545 static bool zbd_is_seq_job(const struct fio_file *f)
547 uint32_t zone_idx, zone_idx_b, zone_idx_e;
554 zone_idx_b = zbd_offset_to_zone_idx(f, f->file_offset);
556 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size - 1);
557 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
558 if (zbd_get_zone(f, zone_idx)->has_wp)
565 * Verify whether the file offset and size parameters are aligned with zone
566 * boundaries. If the file offset is not aligned, align it down to the start of
567 * the zone containing the start offset and align up the file io_size parameter.
569 static bool zbd_zone_align_file_sizes(struct thread_data *td,
572 const struct fio_zone_info *z;
573 uint64_t new_offset, new_end;
577 if (f->file_offset >= f->real_file_size)
579 if (!zbd_is_seq_job(f))
582 if (!td->o.zone_size) {
583 td->o.zone_size = f->zbd_info->zone_size;
584 if (!td->o.zone_size) {
585 log_err("%s: invalid 0 zone size\n",
589 } else if (td->o.zone_size != f->zbd_info->zone_size) {
590 log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n",
591 f->file_name, td->o.zone_size,
592 f->zbd_info->zone_size);
596 if (td->o.zone_skip % td->o.zone_size) {
597 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
598 f->file_name, td->o.zone_skip,
603 z = zbd_offset_to_zone(f, f->file_offset);
604 if ((f->file_offset != z->start) &&
605 (td->o.td_ddir != TD_DDIR_READ)) {
606 new_offset = zbd_zone_end(z);
607 if (new_offset >= f->file_offset + f->io_size) {
608 log_info("%s: io_size must be at least one zone\n",
612 log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n",
613 f->file_name, f->file_offset,
615 f->io_size -= (new_offset - f->file_offset);
616 f->file_offset = new_offset;
619 z = zbd_offset_to_zone(f, f->file_offset + f->io_size);
621 if ((td->o.td_ddir != TD_DDIR_READ) &&
622 (f->file_offset + f->io_size != new_end)) {
623 if (new_end <= f->file_offset) {
624 log_info("%s: io_size must be at least one zone\n",
628 log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n",
629 f->file_name, f->io_size,
630 new_end - f->file_offset);
631 f->io_size = new_end - f->file_offset;
638 * Verify whether offset and size parameters are aligned with zone boundaries.
640 static bool zbd_verify_sizes(void)
642 struct thread_data *td;
647 for_each_file(td, f, j) {
648 if (!zbd_zone_align_file_sizes(td, f))
656 static bool zbd_verify_bs(void)
658 struct thread_data *td;
664 (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] ||
665 td->o.bssplit_nr[DDIR_TRIM])) {
666 log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n");
669 for_each_file(td, f, j) {
675 zone_size = f->zbd_info->zone_size;
676 if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) {
677 log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n",
678 f->file_name, td->o.bs[DDIR_TRIM],
687 static int ilog2(uint64_t i)
699 * Initialize f->zbd_info for devices that are not zoned block devices. This
700 * allows to execute a ZBD workload against a non-ZBD device.
702 static int init_zone_info(struct thread_data *td, struct fio_file *f)
705 struct fio_zone_info *p;
706 uint64_t zone_size = td->o.zone_size;
707 uint64_t zone_capacity = td->o.zone_capacity;
708 struct zoned_block_device_info *zbd_info = NULL;
711 if (zone_size == 0) {
712 log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n",
717 if (zone_size < 512) {
718 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
723 if (zone_capacity == 0)
724 zone_capacity = zone_size;
726 if (zone_capacity > zone_size) {
727 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
728 f->file_name, td->o.zone_capacity, td->o.zone_size);
732 if (f->real_file_size < zone_size) {
733 log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n",
734 f->file_name, f->real_file_size, zone_size);
738 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
739 zbd_info = scalloc(1, sizeof(*zbd_info) +
740 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
744 mutex_init_pshared(&zbd_info->mutex);
745 zbd_info->refcount = 1;
746 p = &zbd_info->zone_info[0];
747 for (i = 0; i < nr_zones; i++, p++) {
748 mutex_init_pshared_with_type(&p->mutex,
749 PTHREAD_MUTEX_RECURSIVE);
750 p->start = i * zone_size;
752 p->type = ZBD_ZONE_TYPE_SWR;
753 p->cond = ZBD_ZONE_COND_EMPTY;
754 p->capacity = zone_capacity;
758 p->start = nr_zones * zone_size;
760 f->zbd_info = zbd_info;
761 f->zbd_info->zone_size = zone_size;
762 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
763 ilog2(zone_size) : 0;
764 f->zbd_info->nr_zones = nr_zones;
769 * Maximum number of zones to report in one operation.
771 #define ZBD_REPORT_MAX_ZONES 8192U
774 * Parse the device zone report and store it in f->zbd_info. Must be called
775 * only for devices that are zoned, namely those with a model != ZBD_NONE.
777 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
780 struct zbd_zone *zones, *z;
781 struct fio_zone_info *p;
782 uint64_t zone_size, offset;
783 struct zoned_block_device_info *zbd_info = NULL;
784 int i, j, ret = -ENOMEM;
786 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
790 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
793 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
798 zone_size = zones[0].len;
799 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
801 if (td->o.zone_size == 0) {
802 td->o.zone_size = zone_size;
803 } else if (td->o.zone_size != zone_size) {
804 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
805 f->file_name, td->o.zone_size, zone_size);
810 dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB and capacity %"PRIu64" KB\n",
811 f->file_name, nr_zones, zone_size / 1024, zones[0].capacity / 1024);
813 zbd_info = scalloc(1, sizeof(*zbd_info) +
814 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
817 mutex_init_pshared(&zbd_info->mutex);
818 zbd_info->refcount = 1;
819 p = &zbd_info->zone_info[0];
820 for (offset = 0, j = 0; j < nr_zones;) {
822 for (i = 0; i < nrz; i++, j++, z++, p++) {
823 mutex_init_pshared_with_type(&p->mutex,
824 PTHREAD_MUTEX_RECURSIVE);
826 p->capacity = z->capacity;
829 case ZBD_ZONE_COND_NOT_WP:
830 case ZBD_ZONE_COND_FULL:
831 p->wp = p->start + p->capacity;
834 assert(z->start <= z->wp);
835 assert(z->wp <= z->start + zone_size);
841 case ZBD_ZONE_TYPE_SWR:
850 if (j > 0 && p->start != p[-1].start + zone_size) {
851 log_info("%s: invalid zone data [%d:%d]: %"PRIu64" + %"PRIu64" != %"PRIu64"\n",
853 p[-1].start, zone_size, p->start);
859 offset = z->start + z->len;
863 nrz = zbd_report_zones(td, f, offset, zones,
864 min((uint32_t)(nr_zones - j),
865 ZBD_REPORT_MAX_ZONES));
868 log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n",
869 offset, f->file_name, -ret);
875 zbd_info->zone_info[nr_zones].start = offset;
877 f->zbd_info = zbd_info;
878 f->zbd_info->zone_size = zone_size;
879 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
880 ilog2(zone_size) : 0;
881 f->zbd_info->nr_zones = nr_zones;
891 static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f)
893 struct zoned_block_device_info *zbd = f->zbd_info;
894 unsigned int max_open_zones;
897 if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) {
898 /* Only host-managed devices have a max open limit */
899 zbd->max_open_zones = td->o.max_open_zones;
903 /* If host-managed, get the max open limit */
904 ret = zbd_get_max_open_zones(td, f, &max_open_zones);
908 if (!max_open_zones) {
909 /* No device limit */
910 zbd->max_open_zones = td->o.max_open_zones;
911 } else if (!td->o.max_open_zones) {
912 /* No user limit. Set limit to device limit */
913 zbd->max_open_zones = max_open_zones;
914 } else if (td->o.max_open_zones <= max_open_zones) {
915 /* Both user limit and dev limit. User limit not too large */
916 zbd->max_open_zones = td->o.max_open_zones;
918 /* Both user limit and dev limit. User limit too large */
919 td_verror(td, EINVAL,
920 "Specified --max_open_zones is too large");
921 log_err("Specified --max_open_zones (%d) is larger than max (%u)\n",
922 td->o.max_open_zones, max_open_zones);
927 /* Ensure that the limit is not larger than FIO's internal limit */
928 if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) {
929 td_verror(td, EINVAL, "'max_open_zones' value is too large");
930 log_err("'max_open_zones' value is larger than %u\n",
935 dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n",
936 f->file_name, zbd->max_open_zones);
942 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
944 * Returns 0 upon success and a negative error code upon failure.
946 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
948 enum zbd_zoned_model zbd_model;
951 assert(td->o.zone_mode == ZONE_MODE_ZBD);
953 ret = zbd_get_zoned_model(td, f, &zbd_model);
959 case ZBD_HOST_MANAGED:
960 ret = parse_zone_info(td, f);
965 ret = init_zone_info(td, f);
970 td_verror(td, EINVAL, "Unsupported zoned model");
971 log_err("Unsupported zoned model\n");
976 f->zbd_info->model = zbd_model;
978 ret = zbd_set_max_open_zones(td, f);
980 zbd_free_zone_info(f);
987 void zbd_free_zone_info(struct fio_file *f)
993 pthread_mutex_lock(&f->zbd_info->mutex);
994 refcount = --f->zbd_info->refcount;
995 pthread_mutex_unlock(&f->zbd_info->mutex);
997 assert((int32_t)refcount >= 0);
1004 * Initialize f->zbd_info.
1006 * Returns 0 upon success and a negative error code upon failure.
1008 * Note: this function can only work correctly if it is called before the first
1011 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
1013 struct thread_data *td2;
1014 struct fio_file *f2;
1017 for_each_td(td2, i) {
1018 for_each_file(td2, f2, j) {
1019 if (td2 == td && f2 == file)
1021 if (!f2->zbd_info ||
1022 strcmp(f2->file_name, file->file_name) != 0)
1024 file->zbd_info = f2->zbd_info;
1025 file->zbd_info->refcount++;
1030 ret = zbd_create_zone_info(td, file);
1032 td_verror(td, -ret, "zbd_create_zone_info() failed");
1037 int zbd_init_files(struct thread_data *td)
1042 for_each_file(td, f, i) {
1043 if (zbd_init_zone_info(td, f))
1050 void zbd_recalc_options_with_zone_granularity(struct thread_data *td)
1055 for_each_file(td, f, i) {
1056 struct zoned_block_device_info *zbd = f->zbd_info;
1059 /* zonemode=strided doesn't get per-file zone size. */
1060 zone_size = zbd ? zbd->zone_size : td->o.zone_size;
1064 if (td->o.size_nz > 0)
1065 td->o.size = td->o.size_nz * zone_size;
1066 if (td->o.io_size_nz > 0)
1067 td->o.io_size = td->o.io_size_nz * zone_size;
1068 if (td->o.start_offset_nz > 0)
1069 td->o.start_offset = td->o.start_offset_nz * zone_size;
1070 if (td->o.offset_increment_nz > 0)
1071 td->o.offset_increment =
1072 td->o.offset_increment_nz * zone_size;
1073 if (td->o.zone_skip_nz > 0)
1074 td->o.zone_skip = td->o.zone_skip_nz * zone_size;
1078 static uint64_t zbd_verify_and_set_vdb(struct thread_data *td,
1079 const struct fio_file *f)
1081 struct fio_zone_info *zb, *ze, *z;
1082 uint64_t wp_vdb = 0;
1083 struct zoned_block_device_info *zbdi = f->zbd_info;
1085 assert(td->runstate < TD_RUNNING);
1088 if (!accounting_vdb(td, f))
1092 * Ensure that the I/O range includes one or more sequential zones so
1093 * that f->min_zone and f->max_zone have different values.
1095 if (!zbd_is_seq_job(f))
1098 if (zbdi->write_min_zone != zbdi->write_max_zone) {
1099 if (zbdi->write_min_zone != f->min_zone ||
1100 zbdi->write_max_zone != f->max_zone) {
1101 td_verror(td, EINVAL,
1102 "multi-jobs with different write ranges are "
1103 "not supported with zone_reset_threshold");
1104 log_err("multi-jobs with different write ranges are "
1105 "not supported with zone_reset_threshold\n");
1110 zbdi->write_min_zone = f->min_zone;
1111 zbdi->write_max_zone = f->max_zone;
1113 zb = zbd_get_zone(f, f->min_zone);
1114 ze = zbd_get_zone(f, f->max_zone);
1115 for (z = zb; z < ze; z++)
1117 wp_vdb += z->wp - z->start;
1119 zbdi->wp_valid_data_bytes = wp_vdb;
1124 int zbd_setup_files(struct thread_data *td)
1129 if (!zbd_using_direct_io()) {
1130 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
1134 if (!zbd_verify_sizes())
1137 if (!zbd_verify_bs())
1140 if (td->o.experimental_verify) {
1141 log_err("zonemode=zbd does not support experimental verify\n");
1145 for_each_file(td, f, i) {
1146 struct zoned_block_device_info *zbd = f->zbd_info;
1147 struct fio_zone_info *z;
1153 f->min_zone = zbd_offset_to_zone_idx(f, f->file_offset);
1155 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size);
1157 vdb = zbd_verify_and_set_vdb(td, f);
1159 dprint(FD_ZBD, "%s(%s): valid data bytes = %" PRIu64 "\n",
1160 __func__, f->file_name, vdb);
1163 * When all zones in the I/O range are conventional, io_size
1164 * can be smaller than zone size, making min_zone the same
1165 * as max_zone. This is why the assert below needs to be made
1168 if (zbd_is_seq_job(f))
1169 assert(f->min_zone < f->max_zone);
1171 if (td->o.max_open_zones > 0 &&
1172 zbd->max_open_zones != td->o.max_open_zones) {
1173 log_err("Different 'max_open_zones' values\n");
1178 * The per job max open zones limit cannot be used without a
1179 * global max open zones limit. (As the tracking of open zones
1180 * is disabled when there is no global max open zones limit.)
1182 if (td->o.job_max_open_zones && !zbd->max_open_zones) {
1183 log_err("'job_max_open_zones' cannot be used without a global open zones limit\n");
1188 * zbd->max_open_zones is the global limit shared for all jobs
1189 * that target the same zoned block device. Force sync the per
1190 * thread global limit with the actual global limit. (The real
1191 * per thread/job limit is stored in td->o.job_max_open_zones).
1193 td->o.max_open_zones = zbd->max_open_zones;
1195 for (zi = f->min_zone; zi < f->max_zone; zi++) {
1196 z = &zbd->zone_info[zi];
1197 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
1198 z->cond != ZBD_ZONE_COND_EXP_OPEN)
1200 if (zbd_open_zone(td, f, z))
1203 * If the number of open zones exceeds specified limits,
1204 * reset all extra open zones.
1206 if (zbd_reset_zone(td, f, z) < 0) {
1207 log_err("Failed to reest zone %d\n", zi);
1217 * Reset zbd_info.write_cnt, the counter that counts down towards the next
1220 static void _zbd_reset_write_cnt(const struct thread_data *td,
1221 const struct fio_file *f)
1223 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
1225 f->zbd_info->write_cnt = td->o.zrf.u.f ?
1226 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
1229 static void zbd_reset_write_cnt(const struct thread_data *td,
1230 const struct fio_file *f)
1232 pthread_mutex_lock(&f->zbd_info->mutex);
1233 _zbd_reset_write_cnt(td, f);
1234 pthread_mutex_unlock(&f->zbd_info->mutex);
1237 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
1238 const struct fio_file *f)
1240 uint32_t write_cnt = 0;
1242 pthread_mutex_lock(&f->zbd_info->mutex);
1243 assert(f->zbd_info->write_cnt);
1244 if (f->zbd_info->write_cnt)
1245 write_cnt = --f->zbd_info->write_cnt;
1247 _zbd_reset_write_cnt(td, f);
1248 pthread_mutex_unlock(&f->zbd_info->mutex);
1250 return write_cnt == 0;
1253 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
1255 struct fio_zone_info *zb, *ze;
1256 bool verify_data_left = false;
1258 if (!f->zbd_info || !td_write(td))
1261 zb = zbd_get_zone(f, f->min_zone);
1262 ze = zbd_get_zone(f, f->max_zone);
1265 * If data verification is enabled reset the affected zones before
1266 * writing any data to avoid that a zone reset has to be issued while
1267 * writing data, which causes data loss.
1269 if (td->o.verify != VERIFY_NONE) {
1270 verify_data_left = td->runstate == TD_VERIFYING ||
1271 td->io_hist_len || td->verify_batch;
1272 if (td->io_hist_len && td->o.verify_backlog)
1274 td->io_hist_len % td->o.verify_backlog;
1275 if (!verify_data_left)
1276 zbd_reset_zones(td, f, zb, ze);
1279 zbd_reset_write_cnt(td, f);
1282 /* Return random zone index for one of the open zones. */
1283 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1284 const struct io_u *io_u)
1286 return (io_u->offset - f->file_offset) *
1287 f->zbd_info->num_open_zones / f->io_size;
1290 static bool any_io_in_flight(void)
1292 struct thread_data *td;
1295 for_each_td(td, i) {
1296 if (td->io_u_in_flight)
1304 * Modify the offset of an I/O unit that does not refer to an open zone such
1305 * that it refers to an open zone. Close an open zone and open a new zone if
1306 * necessary. The open zone is searched across sequential zones.
1307 * This algorithm can only work correctly if all write pointers are
1308 * a multiple of the fio block size. The caller must neither hold z->mutex
1309 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1311 static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
1314 const uint64_t min_bs = td->o.min_bs[io_u->ddir];
1315 struct fio_file *f = io_u->file;
1316 struct zoned_block_device_info *zbdi = f->zbd_info;
1317 struct fio_zone_info *z;
1318 unsigned int open_zone_idx = -1;
1319 uint32_t zone_idx, new_zone_idx;
1321 bool wait_zone_close;
1323 bool should_retry = true;
1325 assert(is_valid_offset(f, io_u->offset));
1327 if (zbdi->max_open_zones || td->o.job_max_open_zones) {
1329 * This statement accesses zbdi->open_zones[] on purpose
1332 zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)];
1334 zone_idx = zbd_offset_to_zone_idx(f, io_u->offset);
1336 if (zone_idx < f->min_zone)
1337 zone_idx = f->min_zone;
1338 else if (zone_idx >= f->max_zone)
1339 zone_idx = f->max_zone - 1;
1342 "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1343 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1346 * Since z->mutex is the outer lock and zbdi->mutex the inner
1347 * lock it can happen that the state of the zone with index zone_idx
1348 * has changed after 'z' has been assigned and before zbdi->mutex
1349 * has been obtained. Hence the loop.
1354 z = zbd_get_zone(f, zone_idx);
1356 zone_lock(td, f, z);
1358 pthread_mutex_lock(&zbdi->mutex);
1361 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1362 zbdi->max_open_zones == 0 &&
1363 td->o.job_max_open_zones == 0)
1365 if (zbdi->num_open_zones == 0) {
1366 dprint(FD_ZBD, "%s(%s): no zones are open\n",
1367 __func__, f->file_name);
1368 goto open_other_zone;
1373 * List of opened zones is per-device, shared across all
1374 * threads. Start with quasi-random candidate zone. Ignore
1375 * zones which don't belong to thread's offset/size area.
1377 open_zone_idx = pick_random_zone_idx(f, io_u);
1378 assert(!open_zone_idx ||
1379 open_zone_idx < zbdi->num_open_zones);
1380 tmp_idx = open_zone_idx;
1382 for (i = 0; i < zbdi->num_open_zones; i++) {
1385 if (tmp_idx >= zbdi->num_open_zones)
1387 tmpz = zbdi->open_zones[tmp_idx];
1388 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1389 open_zone_idx = tmp_idx;
1390 goto found_candidate_zone;
1396 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1397 __func__, f->file_name);
1399 pthread_mutex_unlock(&zbdi->mutex);
1406 found_candidate_zone:
1407 new_zone_idx = zbdi->open_zones[open_zone_idx];
1408 if (new_zone_idx == zone_idx)
1410 zone_idx = new_zone_idx;
1412 pthread_mutex_unlock(&zbdi->mutex);
1418 /* Both z->mutex and zbdi->mutex are held. */
1421 if (zbd_zone_remainder(z) >= min_bs) {
1422 pthread_mutex_unlock(&zbdi->mutex);
1427 /* Check if number of open zones reaches one of limits. */
1429 zbdi->num_open_zones == f->max_zone - f->min_zone ||
1430 (zbdi->max_open_zones &&
1431 zbdi->num_open_zones == zbdi->max_open_zones) ||
1432 (td->o.job_max_open_zones &&
1433 td->num_open_zones == td->o.job_max_open_zones);
1435 pthread_mutex_unlock(&zbdi->mutex);
1437 /* Only z->mutex is held. */
1440 * When number of open zones reaches to one of limits, wait for
1441 * zone close before opening a new zone.
1443 if (wait_zone_close) {
1445 "%s(%s): quiesce to allow open zones to close\n",
1446 __func__, f->file_name);
1451 /* Zone 'z' is full, so try to open a new zone. */
1452 for (i = f->io_size / zbdi->zone_size; i > 0; i--) {
1457 if (!is_valid_offset(f, z->start)) {
1459 zone_idx = f->min_zone;
1460 z = zbd_get_zone(f, zone_idx);
1462 assert(is_valid_offset(f, z->start));
1465 zone_lock(td, f, z);
1468 if (zbd_open_zone(td, f, z))
1472 /* Only z->mutex is held. */
1474 /* Check whether the write fits in any of the already opened zones. */
1475 pthread_mutex_lock(&zbdi->mutex);
1476 for (i = 0; i < zbdi->num_open_zones; i++) {
1477 zone_idx = zbdi->open_zones[i];
1478 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1480 pthread_mutex_unlock(&zbdi->mutex);
1483 z = zbd_get_zone(f, zone_idx);
1485 zone_lock(td, f, z);
1486 if (zbd_zone_remainder(z) >= min_bs)
1488 pthread_mutex_lock(&zbdi->mutex);
1492 * When any I/O is in-flight or when all I/Os in-flight get completed,
1493 * the I/Os might have closed zones then retry the steps to open a zone.
1494 * Before retry, call io_u_quiesce() to complete in-flight writes.
1496 in_flight = any_io_in_flight();
1497 if (in_flight || should_retry) {
1499 "%s(%s): wait zone close and retry open zones\n",
1500 __func__, f->file_name);
1501 pthread_mutex_unlock(&zbdi->mutex);
1504 zone_lock(td, f, z);
1505 should_retry = in_flight;
1509 pthread_mutex_unlock(&zbdi->mutex);
1513 dprint(FD_ZBD, "%s(%s): did not open another zone\n",
1514 __func__, f->file_name);
1519 dprint(FD_ZBD, "%s(%s): returning zone %d\n",
1520 __func__, f->file_name, zone_idx);
1522 io_u->offset = z->start;
1524 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1530 * Find another zone which has @min_bytes of readable data. Search in zones
1531 * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf.
1533 * Either returns NULL or returns a zone pointer. When the zone has write
1534 * pointer, hold the mutex for the zone.
1536 static struct fio_zone_info *
1537 zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes,
1538 struct fio_zone_info *zb, struct fio_zone_info *zl)
1540 struct fio_file *f = io_u->file;
1541 struct fio_zone_info *z1, *z2;
1542 const struct fio_zone_info *const zf = zbd_get_zone(f, f->min_zone);
1545 * Skip to the next non-empty zone in case of sequential I/O and to
1546 * the nearest non-empty zone in case of random I/O.
1548 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1549 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1551 zone_lock(td, f, z1);
1552 if (z1->start + min_bytes <= z1->wp)
1556 } else if (!td_random(td)) {
1560 if (td_random(td) && z2 >= zf &&
1561 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1563 zone_lock(td, f, z2);
1564 if (z2->start + min_bytes <= z2->wp)
1572 "%s: no zone has %"PRIu64" bytes of readable data\n",
1573 f->file_name, min_bytes);
1579 * zbd_end_zone_io - update zone status at command completion
1581 * @z: zone info pointer
1583 * If the write command made the zone full, close it.
1585 * The caller must hold z->mutex.
1587 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1588 struct fio_zone_info *z)
1590 const struct fio_file *f = io_u->file;
1592 if (io_u->ddir == DDIR_WRITE &&
1593 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1594 pthread_mutex_lock(&f->zbd_info->mutex);
1595 zbd_close_zone(td, f, z);
1596 pthread_mutex_unlock(&f->zbd_info->mutex);
1601 * zbd_queue_io - update the write pointer of a sequential zone
1603 * @success: Whether or not the I/O unit has been queued successfully
1604 * @q: queueing status (busy, completed or queued).
1606 * For write and trim operations, update the write pointer of the I/O unit
1609 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1612 const struct fio_file *f = io_u->file;
1613 struct zoned_block_device_info *zbd_info = f->zbd_info;
1614 struct fio_zone_info *z;
1619 z = zbd_offset_to_zone(f, io_u->offset);
1626 "%s: queued I/O (%lld, %llu) for zone %u\n",
1627 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1629 switch (io_u->ddir) {
1631 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1632 zbd_zone_capacity_end(z));
1635 * z->wp > zone_end means that one or more I/O errors
1638 if (accounting_vdb(td, f) && z->wp <= zone_end) {
1639 pthread_mutex_lock(&zbd_info->mutex);
1640 zbd_info->wp_valid_data_bytes += zone_end - z->wp;
1641 pthread_mutex_unlock(&zbd_info->mutex);
1649 if (q == FIO_Q_COMPLETED && !io_u->error)
1650 zbd_end_zone_io(td, io_u, z);
1653 if (!success || q != FIO_Q_QUEUED) {
1654 /* BUSY or COMPLETED: unlock the zone */
1656 io_u->zbd_put_io = NULL;
1661 * zbd_put_io - Unlock an I/O unit target zone lock
1664 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1666 const struct fio_file *f = io_u->file;
1667 struct zoned_block_device_info *zbd_info = f->zbd_info;
1668 struct fio_zone_info *z;
1672 z = zbd_offset_to_zone(f, io_u->offset);
1676 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1677 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1679 zbd_end_zone_io(td, io_u, z);
1685 * Windows and MacOS do not define this.
1688 #define EREMOTEIO 121 /* POSIX value */
1691 bool zbd_unaligned_write(int error_code)
1693 switch (error_code) {
1702 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1703 * @td: FIO thread data.
1704 * @io_u: FIO I/O unit.
1706 * For sequential workloads, change the file offset to skip zoneskip bytes when
1707 * no more IO can be performed in the current zone.
1708 * - For read workloads, zoneskip is applied when the io has reached the end of
1709 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1710 * - For write workloads, zoneskip is applied when the zone is full.
1711 * This applies only to read and write operations.
1713 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1715 struct fio_file *f = io_u->file;
1716 enum fio_ddir ddir = io_u->ddir;
1717 struct fio_zone_info *z;
1719 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1720 assert(td->o.zone_size);
1721 assert(f->zbd_info);
1723 z = zbd_offset_to_zone(f, f->last_pos[ddir]);
1726 * When the zone capacity is smaller than the zone size and the I/O is
1727 * sequential write, skip to zone end if the latest position is at the
1728 * zone capacity limit.
1730 if (z->capacity < f->zbd_info->zone_size &&
1731 !td_random(td) && ddir == DDIR_WRITE &&
1732 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1734 "%s: Jump from zone capacity limit to zone end:"
1735 " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n",
1736 f->file_name, f->last_pos[ddir],
1737 zbd_zone_end(z), zbd_zone_idx(f, z), z->capacity);
1738 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1739 f->last_pos[ddir] = zbd_zone_end(z);
1743 * zone_skip is valid only for sequential workloads.
1745 if (td_random(td) || !td->o.zone_skip)
1749 * It is time to switch to a new zone if:
1750 * - zone_bytes == zone_size bytes have already been accessed
1751 * - The last position reached the end of the current zone.
1752 * - For reads with td->o.read_beyond_wp == false, the last position
1753 * reached the zone write pointer.
1755 if (td->zone_bytes >= td->o.zone_size ||
1756 f->last_pos[ddir] >= zbd_zone_end(z) ||
1757 (ddir == DDIR_READ &&
1758 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1763 f->file_offset += td->o.zone_size + td->o.zone_skip;
1766 * Wrap from the beginning, if we exceed the file size
1768 if (f->file_offset >= f->real_file_size)
1769 f->file_offset = get_start_offset(td, f);
1771 f->last_pos[ddir] = f->file_offset;
1772 td->io_skip_bytes += td->o.zone_skip;
1777 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1779 * @td: FIO thread data.
1780 * @io_u: FIO I/O unit.
1781 * @ddir: I/O direction before adjustment.
1783 * Return adjusted I/O direction.
1785 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1789 * In case read direction is chosen for the first random I/O, fio with
1790 * zonemode=zbd stops because no data can be read from zoned block
1791 * devices with all empty zones. Overwrite the first I/O direction as
1792 * write to make sure data to read exists.
1794 assert(io_u->file->zbd_info);
1795 if (ddir != DDIR_READ || !td_rw(td))
1798 if (io_u->file->last_start[DDIR_WRITE] != -1ULL || td->o.read_beyond_wp)
1805 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1806 * @td: FIO thread data.
1807 * @io_u: FIO I/O unit.
1809 * Locking strategy: returns with z->mutex locked if and only if z refers
1810 * to a sequential zone and if io_u_accept is returned. z is the zone that
1811 * corresponds to io_u->offset at the end of this function.
1813 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1815 struct fio_file *f = io_u->file;
1816 struct zoned_block_device_info *zbdi = f->zbd_info;
1817 struct fio_zone_info *zb, *zl, *orig_zb;
1818 uint32_t orig_len = io_u->buflen;
1819 uint64_t min_bs = td->o.min_bs[io_u->ddir];
1825 assert(is_valid_offset(f, io_u->offset));
1826 assert(io_u->buflen);
1828 zb = zbd_offset_to_zone(f, io_u->offset);
1832 /* Accept non-write I/Os for conventional zones. */
1833 if (io_u->ddir != DDIR_WRITE)
1837 * Make sure that writes to conventional zones
1838 * don't cross over to any sequential zones.
1840 if (!(zb + 1)->has_wp ||
1841 io_u->offset + io_u->buflen <= (zb + 1)->start)
1844 if (io_u->offset + min_bs > (zb + 1)->start) {
1846 "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n",
1847 f->file_name, io_u->offset,
1848 min_bs, (zb + 1)->start);
1850 zb->start + (zb + 1)->start - io_u->offset;
1851 new_len = min(io_u->buflen,
1852 (zb + 1)->start - io_u->offset);
1854 new_len = (zb + 1)->start - io_u->offset;
1857 io_u->buflen = new_len / min_bs * min_bs;
1863 * Accept the I/O offset for reads if reading beyond the write pointer
1866 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1867 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1870 zone_lock(td, f, zb);
1872 switch (io_u->ddir) {
1874 if (td->runstate == TD_VERIFYING && td_write(td))
1878 * Check that there is enough written data in the zone to do an
1879 * I/O of at least min_bs B. If there isn't, find a new zone for
1882 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1883 zb->wp - zb->start : 0;
1884 if (range < min_bs ||
1885 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1887 zl = zbd_get_zone(f, f->max_zone);
1888 zb = zbd_find_zone(td, io_u, min_bs, zb, zl);
1891 "%s: zbd_find_zone(%lld, %llu) failed\n",
1892 f->file_name, io_u->offset,
1897 * zbd_find_zone() returned a zone with a range of at
1900 range = zb->wp - zb->start;
1901 assert(range >= min_bs);
1904 io_u->offset = zb->start;
1908 * Make sure the I/O is within the zone valid data range while
1909 * maximizing the I/O size and preserving randomness.
1911 if (range <= io_u->buflen)
1912 io_u->offset = zb->start;
1913 else if (td_random(td))
1914 io_u->offset = zb->start +
1915 ((io_u->offset - orig_zb->start) %
1916 (range - io_u->buflen)) / min_bs * min_bs;
1919 * When zbd_find_zone() returns a conventional zone,
1920 * we can simply accept the new i/o offset here.
1926 * Make sure the I/O does not cross over the zone wp position.
1928 new_len = min((unsigned long long)io_u->buflen,
1929 (unsigned long long)(zb->wp - io_u->offset));
1930 new_len = new_len / min_bs * min_bs;
1931 if (new_len < io_u->buflen) {
1932 io_u->buflen = new_len;
1933 dprint(FD_IO, "Changed length from %u into %llu\n",
1934 orig_len, io_u->buflen);
1937 assert(zb->start <= io_u->offset);
1938 assert(io_u->offset + io_u->buflen <= zb->wp);
1943 if (io_u->buflen > zbdi->zone_size) {
1944 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
1946 "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n",
1947 f->file_name, io_u->buflen, zbdi->zone_size);
1952 if (zbd_zone_remainder(zb) > 0 &&
1953 zbd_zone_remainder(zb) < min_bs) {
1954 pthread_mutex_lock(&f->zbd_info->mutex);
1955 zbd_close_zone(td, f, zb);
1956 pthread_mutex_unlock(&f->zbd_info->mutex);
1958 "%s: finish zone %d\n",
1959 f->file_name, zbd_zone_idx(f, zb));
1961 zbd_finish_zone(td, f, zb);
1962 if (zbd_zone_idx(f, zb) + 1 >= f->max_zone) {
1968 /* Find the next write pointer zone */
1971 if (zbd_zone_idx(f, zb) >= f->max_zone)
1972 zb = zbd_get_zone(f, f->min_zone);
1973 } while (!zb->has_wp);
1975 zone_lock(td, f, zb);
1978 if (!zbd_open_zone(td, f, zb)) {
1980 zb = zbd_convert_to_open_zone(td, io_u);
1982 dprint(FD_IO, "%s: can't convert to open zone",
1988 if (zbd_zone_remainder(zb) > 0 &&
1989 zbd_zone_remainder(zb) < min_bs)
1992 /* Check whether the zone reset threshold has been exceeded */
1993 if (td->o.zrf.u.f) {
1994 if (zbdi->wp_valid_data_bytes >=
1995 f->io_size * td->o.zrt.u.f &&
1996 zbd_dec_and_reset_write_cnt(td, f))
2000 /* Reset the zone pointer if necessary */
2001 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
2002 if (td->o.verify != VERIFY_NONE) {
2004 * Unset io-u->file to tell get_next_verify()
2005 * that this IO is not requeue.
2008 if (!get_next_verify(td, io_u)) {
2016 * Since previous write requests may have been submitted
2017 * asynchronously and since we will submit the zone
2018 * reset synchronously, wait until previously submitted
2019 * write requests have completed before issuing a
2024 if (zbd_reset_zone(td, f, zb) < 0)
2027 if (zb->capacity < min_bs) {
2028 td_verror(td, EINVAL, "ZCAP is less min_bs");
2029 log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n",
2030 zb->capacity, min_bs);
2035 /* Make writes occur at the write pointer */
2036 assert(!zbd_zone_full(f, zb, min_bs));
2037 io_u->offset = zb->wp;
2038 if (!is_valid_offset(f, io_u->offset)) {
2039 td_verror(td, EINVAL, "invalid WP value");
2040 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
2041 f->file_name, io_u->offset);
2046 * Make sure that the buflen is a multiple of the minimal
2047 * block size. Give up if shrinking would make the request too
2050 new_len = min((unsigned long long)io_u->buflen,
2051 zbd_zone_capacity_end(zb) - io_u->offset);
2052 new_len = new_len / min_bs * min_bs;
2053 if (new_len == io_u->buflen)
2055 if (new_len >= min_bs) {
2056 io_u->buflen = new_len;
2057 dprint(FD_IO, "Changed length from %u into %llu\n",
2058 orig_len, io_u->buflen);
2062 td_verror(td, EIO, "zone remainder too small");
2063 log_err("zone remainder %lld smaller than min block size %"PRIu64"\n",
2064 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
2069 /* Check random trim targets a non-empty zone */
2070 if (!td_random(td) || zb->wp > zb->start)
2073 /* Find out a non-empty zone to trim */
2075 zl = zbd_get_zone(f, f->max_zone);
2076 zb = zbd_find_zone(td, io_u, 1, zb, zl);
2078 io_u->offset = zb->start;
2079 dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n",
2080 f->file_name, io_u->offset);
2089 case DDIR_SYNC_FILE_RANGE:
2100 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
2101 assert(!io_u->zbd_queue_io);
2102 assert(!io_u->zbd_put_io);
2104 io_u->zbd_queue_io = zbd_queue_io;
2105 io_u->zbd_put_io = zbd_put_io;
2108 * Since we return with the zone lock still held,
2109 * add an annotation to let Coverity know that it
2112 /* coverity[missing_unlock] */
2117 if (zb && zb->has_wp)
2123 /* Return a string with ZBD statistics */
2124 char *zbd_write_status(const struct thread_stat *ts)
2128 if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0)
2134 * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim
2136 * @td: FIO thread data.
2137 * @io_u: FIO I/O unit.
2139 * It is assumed that z->mutex is already locked.
2140 * Return io_u_completed when reset zone succeeds. Return 0 when the target zone
2141 * does not have write pointer. On error, return negative errno.
2143 int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u)
2145 struct fio_file *f = io_u->file;
2146 struct fio_zone_info *z;
2149 z = zbd_offset_to_zone(f, io_u->offset);
2153 if (io_u->offset != z->start) {
2154 log_err("Trim offset not at zone start (%lld)\n",
2159 ret = zbd_reset_zone((struct thread_data *)td, f, z);
2163 return io_u_completed;