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));
151 * zbd_get_zoned_model - Get a device zoned model
152 * @td: FIO thread data
153 * @f: FIO file for which to get model information
155 static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f,
156 enum zbd_zoned_model *model)
160 if (f->filetype == FIO_TYPE_PIPE) {
161 log_err("zonemode=zbd does not support pipes\n");
165 /* If regular file, always emulate zones inside the file. */
166 if (f->filetype == FIO_TYPE_FILE) {
171 if (td->io_ops && td->io_ops->get_zoned_model)
172 ret = td->io_ops->get_zoned_model(td, f, model);
174 ret = blkzoned_get_zoned_model(td, f, model);
176 td_verror(td, errno, "get zoned model failed");
177 log_err("%s: get zoned model failed (%d).\n",
178 f->file_name, errno);
185 * zbd_report_zones - Get zone information
186 * @td: FIO thread data.
187 * @f: FIO file for which to get zone information
188 * @offset: offset from which to report zones
189 * @zones: Array of struct zbd_zone
190 * @nr_zones: Size of @zones array
192 * Get zone information into @zones starting from the zone at offset @offset
193 * for the device specified by @f.
195 * Returns the number of zones reported upon success and a negative error code
196 * upon failure. If the zone report is empty, always assume an error (device
197 * problem) and return -EIO.
199 static int zbd_report_zones(struct thread_data *td, struct fio_file *f,
200 uint64_t offset, struct zbd_zone *zones,
201 unsigned int nr_zones)
205 if (td->io_ops && td->io_ops->report_zones)
206 ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones);
208 ret = blkzoned_report_zones(td, f, offset, zones, nr_zones);
210 td_verror(td, errno, "report zones failed");
211 log_err("%s: report zones from sector %"PRIu64" failed (%d).\n",
212 f->file_name, offset >> 9, errno);
213 } else if (ret == 0) {
214 td_verror(td, errno, "Empty zone report");
215 log_err("%s: report zones from sector %"PRIu64" is empty.\n",
216 f->file_name, offset >> 9);
224 * zbd_reset_wp - reset the write pointer of a range of zones
225 * @td: FIO thread data.
226 * @f: FIO file for which to reset zones
227 * @offset: Starting offset of the first zone to reset
228 * @length: Length of the range of zones to reset
230 * Reset the write pointer of all zones in the range @offset...@offset+@length.
231 * Returns 0 upon success and a negative error code upon failure.
233 static int zbd_reset_wp(struct thread_data *td, struct fio_file *f,
234 uint64_t offset, uint64_t length)
238 if (td->io_ops && td->io_ops->reset_wp)
239 ret = td->io_ops->reset_wp(td, f, offset, length);
241 ret = blkzoned_reset_wp(td, f, offset, length);
243 td_verror(td, errno, "resetting wp failed");
244 log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n",
245 f->file_name, length >> 9, offset >> 9, errno);
252 * zbd_reset_zone - reset the write pointer of a single zone
253 * @td: FIO thread data.
254 * @f: FIO file associated with the disk for which to reset a write pointer.
257 * Returns 0 upon success and a negative error code upon failure.
259 * The caller must hold z->mutex.
261 static int zbd_reset_zone(struct thread_data *td, struct fio_file *f,
262 struct fio_zone_info *z)
264 uint64_t offset = z->start;
265 uint64_t length = (z+1)->start - offset;
266 uint64_t data_in_zone = z->wp - z->start;
272 assert(is_valid_offset(f, offset + length - 1));
274 dprint(FD_ZBD, "%s: resetting wp of zone %u.\n",
275 f->file_name, zbd_zone_idx(f, z));
277 switch (f->zbd_info->model) {
279 case ZBD_HOST_MANAGED:
280 ret = zbd_reset_wp(td, f, offset, length);
288 pthread_mutex_lock(&f->zbd_info->mutex);
289 f->zbd_info->wp_valid_data_bytes -= data_in_zone;
290 pthread_mutex_unlock(&f->zbd_info->mutex);
294 td->ts.nr_zone_resets++;
300 * zbd_close_zone - Remove a zone from the open zones array.
301 * @td: FIO thread data.
302 * @f: FIO file associated with the disk for which to reset a write pointer.
303 * @zone_idx: Index of the zone to remove.
305 * The caller must hold f->zbd_info->mutex.
307 static void zbd_close_zone(struct thread_data *td, const struct fio_file *f,
308 struct fio_zone_info *z)
315 for (ozi = 0; ozi < f->zbd_info->num_open_zones; ozi++) {
316 if (zbd_get_zone(f, f->zbd_info->open_zones[ozi]) == z)
319 if (ozi == f->zbd_info->num_open_zones)
322 dprint(FD_ZBD, "%s: closing zone %u\n",
323 f->file_name, zbd_zone_idx(f, z));
325 memmove(f->zbd_info->open_zones + ozi,
326 f->zbd_info->open_zones + ozi + 1,
327 (ZBD_MAX_OPEN_ZONES - (ozi + 1)) *
328 sizeof(f->zbd_info->open_zones[0]));
330 f->zbd_info->num_open_zones--;
331 td->num_open_zones--;
336 * zbd_finish_zone - finish the specified zone
337 * @td: FIO thread data.
338 * @f: FIO file for which to finish a zone
339 * @z: Zone to finish.
341 * Finish the zone at @offset with open or close status.
343 static int zbd_finish_zone(struct thread_data *td, struct fio_file *f,
344 struct fio_zone_info *z)
346 uint64_t offset = z->start;
347 uint64_t length = f->zbd_info->zone_size;
350 switch (f->zbd_info->model) {
352 case ZBD_HOST_MANAGED:
353 if (td->io_ops && td->io_ops->finish_zone)
354 ret = td->io_ops->finish_zone(td, f, offset, length);
356 ret = blkzoned_finish_zone(td, f, offset, length);
363 td_verror(td, errno, "finish zone failed");
364 log_err("%s: finish zone at sector %"PRIu64" failed (%d).\n",
365 f->file_name, offset >> 9, errno);
367 z->wp = (z+1)->start;
374 * zbd_reset_zones - Reset a range of zones.
375 * @td: fio thread data.
376 * @f: fio file for which to reset zones
377 * @zb: first zone to reset.
378 * @ze: first zone not to reset.
380 * Returns 0 upon success and 1 upon failure.
382 static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
383 struct fio_zone_info *const zb,
384 struct fio_zone_info *const ze)
386 struct fio_zone_info *z;
387 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
392 dprint(FD_ZBD, "%s: examining zones %u .. %u\n",
393 f->file_name, zbd_zone_idx(f, zb), zbd_zone_idx(f, ze));
395 for (z = zb; z < ze; z++) {
400 pthread_mutex_lock(&f->zbd_info->mutex);
401 zbd_close_zone(td, f, z);
402 pthread_mutex_unlock(&f->zbd_info->mutex);
404 if (z->wp != z->start) {
405 dprint(FD_ZBD, "%s: resetting zone %u\n",
406 f->file_name, zbd_zone_idx(f, z));
407 if (zbd_reset_zone(td, f, z) < 0)
418 * zbd_get_max_open_zones - Get the maximum number of open zones
419 * @td: FIO thread data
420 * @f: FIO file for which to get max open zones
421 * @max_open_zones: Upon success, result will be stored here.
423 * A @max_open_zones value set to zero means no limit.
425 * Returns 0 upon success and a negative error code upon failure.
427 static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f,
428 unsigned int *max_open_zones)
432 if (td->io_ops && td->io_ops->get_max_open_zones)
433 ret = td->io_ops->get_max_open_zones(td, f, max_open_zones);
435 ret = blkzoned_get_max_open_zones(td, f, max_open_zones);
437 td_verror(td, errno, "get max open zones failed");
438 log_err("%s: get max open zones failed (%d).\n",
439 f->file_name, errno);
446 * zbd_open_zone - Add a zone to the array of open zones.
447 * @td: fio thread data.
448 * @f: fio file that has the open zones to add.
449 * @zone_idx: Index of the zone to add.
451 * Open a ZBD zone if it is not already open. Returns true if either the zone
452 * was already open or if the zone was successfully added to the array of open
453 * zones without exceeding the maximum number of open zones. Returns false if
454 * the zone was not already open and opening the zone would cause the zone limit
457 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
458 struct fio_zone_info *z)
460 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
461 struct zoned_block_device_info *zbdi = f->zbd_info;
462 uint32_t zone_idx = zbd_zone_idx(f, z);
465 if (z->cond == ZBD_ZONE_COND_OFFLINE)
469 * Skip full zones with data verification enabled because resetting a
470 * zone causes data loss and hence causes verification to fail.
472 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
476 * zbdi->max_open_zones == 0 means that there is no limit on the maximum
477 * number of open zones. In this case, do no track open zones in
478 * zbdi->open_zones array.
480 if (!zbdi->max_open_zones)
483 pthread_mutex_lock(&zbdi->mutex);
487 * If the zone is going to be completely filled by writes
488 * already in-flight, handle it as a full zone instead of an
491 if (!zbd_zone_remainder(z))
497 /* Zero means no limit */
498 if (td->o.job_max_open_zones > 0 &&
499 td->num_open_zones >= td->o.job_max_open_zones)
501 if (zbdi->num_open_zones >= zbdi->max_open_zones)
504 dprint(FD_ZBD, "%s: opening zone %u\n",
505 f->file_name, zone_idx);
507 zbdi->open_zones[zbdi->num_open_zones++] = zone_idx;
508 td->num_open_zones++;
513 pthread_mutex_unlock(&zbdi->mutex);
517 /* Verify whether direct I/O is used for all host-managed zoned block drives. */
518 static bool zbd_using_direct_io(void)
520 struct thread_data *td;
525 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
527 for_each_file(td, f, j) {
528 if (f->zbd_info && f->filetype == FIO_TYPE_BLOCK &&
529 f->zbd_info->model == ZBD_HOST_MANAGED)
537 /* Whether or not the I/O range for f includes one or more sequential zones */
538 static bool zbd_is_seq_job(struct fio_file *f)
540 uint32_t zone_idx, zone_idx_b, zone_idx_e;
547 zone_idx_b = zbd_offset_to_zone_idx(f, f->file_offset);
549 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size - 1);
550 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
551 if (zbd_get_zone(f, zone_idx)->has_wp)
558 * Verify whether the file offset and size parameters are aligned with zone
559 * boundaries. If the file offset is not aligned, align it down to the start of
560 * the zone containing the start offset and align up the file io_size parameter.
562 static bool zbd_zone_align_file_sizes(struct thread_data *td,
565 const struct fio_zone_info *z;
566 uint64_t new_offset, new_end;
570 if (f->file_offset >= f->real_file_size)
572 if (!zbd_is_seq_job(f))
575 if (!td->o.zone_size) {
576 td->o.zone_size = f->zbd_info->zone_size;
577 if (!td->o.zone_size) {
578 log_err("%s: invalid 0 zone size\n",
582 } else if (td->o.zone_size != f->zbd_info->zone_size) {
583 log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n",
584 f->file_name, td->o.zone_size,
585 f->zbd_info->zone_size);
589 if (td->o.zone_skip % td->o.zone_size) {
590 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
591 f->file_name, td->o.zone_skip,
596 z = zbd_offset_to_zone(f, f->file_offset);
597 if ((f->file_offset != z->start) &&
598 (td->o.td_ddir != TD_DDIR_READ)) {
599 new_offset = zbd_zone_end(z);
600 if (new_offset >= f->file_offset + f->io_size) {
601 log_info("%s: io_size must be at least one zone\n",
605 log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n",
606 f->file_name, f->file_offset,
608 f->io_size -= (new_offset - f->file_offset);
609 f->file_offset = new_offset;
612 z = zbd_offset_to_zone(f, f->file_offset + f->io_size);
614 if ((td->o.td_ddir != TD_DDIR_READ) &&
615 (f->file_offset + f->io_size != new_end)) {
616 if (new_end <= f->file_offset) {
617 log_info("%s: io_size must be at least one zone\n",
621 log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n",
622 f->file_name, f->io_size,
623 new_end - f->file_offset);
624 f->io_size = new_end - f->file_offset;
631 * Verify whether offset and size parameters are aligned with zone boundaries.
633 static bool zbd_verify_sizes(void)
635 struct thread_data *td;
640 for_each_file(td, f, j) {
641 if (!zbd_zone_align_file_sizes(td, f))
649 static bool zbd_verify_bs(void)
651 struct thread_data *td;
657 (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] ||
658 td->o.bssplit_nr[DDIR_TRIM])) {
659 log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n");
662 for_each_file(td, f, j) {
668 zone_size = f->zbd_info->zone_size;
669 if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) {
670 log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n",
671 f->file_name, td->o.bs[DDIR_TRIM],
680 static int ilog2(uint64_t i)
692 * Initialize f->zbd_info for devices that are not zoned block devices. This
693 * allows to execute a ZBD workload against a non-ZBD device.
695 static int init_zone_info(struct thread_data *td, struct fio_file *f)
698 struct fio_zone_info *p;
699 uint64_t zone_size = td->o.zone_size;
700 uint64_t zone_capacity = td->o.zone_capacity;
701 struct zoned_block_device_info *zbd_info = NULL;
704 if (zone_size == 0) {
705 log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n",
710 if (zone_size < 512) {
711 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
716 if (zone_capacity == 0)
717 zone_capacity = zone_size;
719 if (zone_capacity > zone_size) {
720 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
721 f->file_name, td->o.zone_capacity, td->o.zone_size);
725 if (f->real_file_size < zone_size) {
726 log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n",
727 f->file_name, f->real_file_size, zone_size);
731 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
732 zbd_info = scalloc(1, sizeof(*zbd_info) +
733 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
737 mutex_init_pshared(&zbd_info->mutex);
738 zbd_info->refcount = 1;
739 p = &zbd_info->zone_info[0];
740 for (i = 0; i < nr_zones; i++, p++) {
741 mutex_init_pshared_with_type(&p->mutex,
742 PTHREAD_MUTEX_RECURSIVE);
743 p->start = i * zone_size;
745 p->type = ZBD_ZONE_TYPE_SWR;
746 p->cond = ZBD_ZONE_COND_EMPTY;
747 p->capacity = zone_capacity;
751 p->start = nr_zones * zone_size;
753 f->zbd_info = zbd_info;
754 f->zbd_info->zone_size = zone_size;
755 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
756 ilog2(zone_size) : 0;
757 f->zbd_info->nr_zones = nr_zones;
762 * Maximum number of zones to report in one operation.
764 #define ZBD_REPORT_MAX_ZONES 8192U
767 * Parse the device zone report and store it in f->zbd_info. Must be called
768 * only for devices that are zoned, namely those with a model != ZBD_NONE.
770 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
773 struct zbd_zone *zones, *z;
774 struct fio_zone_info *p;
775 uint64_t zone_size, offset;
776 struct zoned_block_device_info *zbd_info = NULL;
777 int i, j, ret = -ENOMEM;
779 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
783 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
786 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
791 zone_size = zones[0].len;
792 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
794 if (td->o.zone_size == 0) {
795 td->o.zone_size = zone_size;
796 } else if (td->o.zone_size != zone_size) {
797 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
798 f->file_name, td->o.zone_size, zone_size);
803 dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n",
804 f->file_name, nr_zones, zone_size / 1024);
806 zbd_info = scalloc(1, sizeof(*zbd_info) +
807 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
810 mutex_init_pshared(&zbd_info->mutex);
811 zbd_info->refcount = 1;
812 p = &zbd_info->zone_info[0];
813 for (offset = 0, j = 0; j < nr_zones;) {
815 for (i = 0; i < nrz; i++, j++, z++, p++) {
816 mutex_init_pshared_with_type(&p->mutex,
817 PTHREAD_MUTEX_RECURSIVE);
819 p->capacity = z->capacity;
822 case ZBD_ZONE_COND_NOT_WP:
823 case ZBD_ZONE_COND_FULL:
824 p->wp = p->start + p->capacity;
827 assert(z->start <= z->wp);
828 assert(z->wp <= z->start + zone_size);
834 case ZBD_ZONE_TYPE_SWR:
843 if (j > 0 && p->start != p[-1].start + zone_size) {
844 log_info("%s: invalid zone data\n",
851 offset = z->start + z->len;
855 nrz = zbd_report_zones(td, f, offset, zones,
856 min((uint32_t)(nr_zones - j),
857 ZBD_REPORT_MAX_ZONES));
860 log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n",
861 offset, f->file_name, -ret);
867 zbd_info->zone_info[nr_zones].start = offset;
869 f->zbd_info = zbd_info;
870 f->zbd_info->zone_size = zone_size;
871 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
872 ilog2(zone_size) : 0;
873 f->zbd_info->nr_zones = nr_zones;
883 static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f)
885 struct zoned_block_device_info *zbd = f->zbd_info;
886 unsigned int max_open_zones;
889 if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) {
890 /* Only host-managed devices have a max open limit */
891 zbd->max_open_zones = td->o.max_open_zones;
895 /* If host-managed, get the max open limit */
896 ret = zbd_get_max_open_zones(td, f, &max_open_zones);
900 if (!max_open_zones) {
901 /* No device limit */
902 zbd->max_open_zones = td->o.max_open_zones;
903 } else if (!td->o.max_open_zones) {
904 /* No user limit. Set limit to device limit */
905 zbd->max_open_zones = max_open_zones;
906 } else if (td->o.max_open_zones <= max_open_zones) {
907 /* Both user limit and dev limit. User limit not too large */
908 zbd->max_open_zones = td->o.max_open_zones;
910 /* Both user limit and dev limit. User limit too large */
911 td_verror(td, EINVAL,
912 "Specified --max_open_zones is too large");
913 log_err("Specified --max_open_zones (%d) is larger than max (%u)\n",
914 td->o.max_open_zones, max_open_zones);
919 /* Ensure that the limit is not larger than FIO's internal limit */
920 if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) {
921 td_verror(td, EINVAL, "'max_open_zones' value is too large");
922 log_err("'max_open_zones' value is larger than %u\n",
927 dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n",
928 f->file_name, zbd->max_open_zones);
934 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
936 * Returns 0 upon success and a negative error code upon failure.
938 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
940 enum zbd_zoned_model zbd_model;
943 assert(td->o.zone_mode == ZONE_MODE_ZBD);
945 ret = zbd_get_zoned_model(td, f, &zbd_model);
951 case ZBD_HOST_MANAGED:
952 ret = parse_zone_info(td, f);
957 ret = init_zone_info(td, f);
962 td_verror(td, EINVAL, "Unsupported zoned model");
963 log_err("Unsupported zoned model\n");
968 f->zbd_info->model = zbd_model;
970 ret = zbd_set_max_open_zones(td, f);
972 zbd_free_zone_info(f);
979 void zbd_free_zone_info(struct fio_file *f)
985 pthread_mutex_lock(&f->zbd_info->mutex);
986 refcount = --f->zbd_info->refcount;
987 pthread_mutex_unlock(&f->zbd_info->mutex);
989 assert((int32_t)refcount >= 0);
996 * Initialize f->zbd_info.
998 * Returns 0 upon success and a negative error code upon failure.
1000 * Note: this function can only work correctly if it is called before the first
1003 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
1005 struct thread_data *td2;
1006 struct fio_file *f2;
1009 for_each_td(td2, i) {
1010 for_each_file(td2, f2, j) {
1011 if (td2 == td && f2 == file)
1013 if (!f2->zbd_info ||
1014 strcmp(f2->file_name, file->file_name) != 0)
1016 file->zbd_info = f2->zbd_info;
1017 file->zbd_info->refcount++;
1022 ret = zbd_create_zone_info(td, file);
1024 td_verror(td, -ret, "zbd_create_zone_info() failed");
1029 int zbd_init_files(struct thread_data *td)
1034 for_each_file(td, f, i) {
1035 if (zbd_init_zone_info(td, f))
1042 void zbd_recalc_options_with_zone_granularity(struct thread_data *td)
1047 for_each_file(td, f, i) {
1048 struct zoned_block_device_info *zbd = f->zbd_info;
1051 /* zonemode=strided doesn't get per-file zone size. */
1052 zone_size = zbd ? zbd->zone_size : td->o.zone_size;
1056 if (td->o.size_nz > 0)
1057 td->o.size = td->o.size_nz * zone_size;
1058 if (td->o.io_size_nz > 0)
1059 td->o.io_size = td->o.io_size_nz * zone_size;
1060 if (td->o.start_offset_nz > 0)
1061 td->o.start_offset = td->o.start_offset_nz * zone_size;
1062 if (td->o.offset_increment_nz > 0)
1063 td->o.offset_increment =
1064 td->o.offset_increment_nz * zone_size;
1065 if (td->o.zone_skip_nz > 0)
1066 td->o.zone_skip = td->o.zone_skip_nz * zone_size;
1070 int zbd_setup_files(struct thread_data *td)
1075 if (!zbd_using_direct_io()) {
1076 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
1080 if (!zbd_verify_sizes())
1083 if (!zbd_verify_bs())
1086 if (td->o.experimental_verify) {
1087 log_err("zonemode=zbd does not support experimental verify\n");
1091 for_each_file(td, f, i) {
1092 struct zoned_block_device_info *zbd = f->zbd_info;
1093 struct fio_zone_info *z;
1098 f->min_zone = zbd_offset_to_zone_idx(f, f->file_offset);
1100 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size);
1103 * When all zones in the I/O range are conventional, io_size
1104 * can be smaller than zone size, making min_zone the same
1105 * as max_zone. This is why the assert below needs to be made
1108 if (zbd_is_seq_job(f))
1109 assert(f->min_zone < f->max_zone);
1111 if (td->o.max_open_zones > 0 &&
1112 zbd->max_open_zones != td->o.max_open_zones) {
1113 log_err("Different 'max_open_zones' values\n");
1118 * The per job max open zones limit cannot be used without a
1119 * global max open zones limit. (As the tracking of open zones
1120 * is disabled when there is no global max open zones limit.)
1122 if (td->o.job_max_open_zones && !zbd->max_open_zones) {
1123 log_err("'job_max_open_zones' cannot be used without a global open zones limit\n");
1128 * zbd->max_open_zones is the global limit shared for all jobs
1129 * that target the same zoned block device. Force sync the per
1130 * thread global limit with the actual global limit. (The real
1131 * per thread/job limit is stored in td->o.job_max_open_zones).
1133 td->o.max_open_zones = zbd->max_open_zones;
1135 for (zi = f->min_zone; zi < f->max_zone; zi++) {
1136 z = &zbd->zone_info[zi];
1137 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
1138 z->cond != ZBD_ZONE_COND_EXP_OPEN)
1140 if (zbd_open_zone(td, f, z))
1143 * If the number of open zones exceeds specified limits,
1144 * reset all extra open zones.
1146 if (zbd_reset_zone(td, f, z) < 0) {
1147 log_err("Failed to reest zone %d\n", zi);
1157 * Reset zbd_info.write_cnt, the counter that counts down towards the next
1160 static void _zbd_reset_write_cnt(const struct thread_data *td,
1161 const struct fio_file *f)
1163 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
1165 f->zbd_info->write_cnt = td->o.zrf.u.f ?
1166 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
1169 static void zbd_reset_write_cnt(const struct thread_data *td,
1170 const struct fio_file *f)
1172 pthread_mutex_lock(&f->zbd_info->mutex);
1173 _zbd_reset_write_cnt(td, f);
1174 pthread_mutex_unlock(&f->zbd_info->mutex);
1177 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
1178 const struct fio_file *f)
1180 uint32_t write_cnt = 0;
1182 pthread_mutex_lock(&f->zbd_info->mutex);
1183 assert(f->zbd_info->write_cnt);
1184 if (f->zbd_info->write_cnt)
1185 write_cnt = --f->zbd_info->write_cnt;
1187 _zbd_reset_write_cnt(td, f);
1188 pthread_mutex_unlock(&f->zbd_info->mutex);
1190 return write_cnt == 0;
1193 static uint64_t zbd_set_vdb(struct thread_data *td, const struct fio_file *f)
1195 struct fio_zone_info *zb, *ze, *z;
1196 uint64_t wp_vdb = 0;
1198 zb = zbd_get_zone(f, f->min_zone);
1199 ze = zbd_get_zone(f, f->max_zone);
1200 for (z = zb; z < ze; z++) {
1202 zone_lock(td, f, z);
1203 wp_vdb += z->wp - z->start;
1207 pthread_mutex_lock(&f->zbd_info->mutex);
1208 f->zbd_info->wp_valid_data_bytes = wp_vdb;
1209 pthread_mutex_unlock(&f->zbd_info->mutex);
1211 for (z = zb; z < ze; z++)
1218 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
1220 struct fio_zone_info *zb, *ze;
1222 bool verify_data_left = false;
1224 if (!f->zbd_info || !td_write(td))
1227 zb = zbd_get_zone(f, f->min_zone);
1228 ze = zbd_get_zone(f, f->max_zone);
1229 vdb = zbd_set_vdb(td, f);
1231 dprint(FD_ZBD, "%s(%s): valid data bytes = %" PRIu64 "\n",
1232 __func__, f->file_name, vdb);
1235 * If data verification is enabled reset the affected zones before
1236 * writing any data to avoid that a zone reset has to be issued while
1237 * writing data, which causes data loss.
1239 if (td->o.verify != VERIFY_NONE) {
1240 verify_data_left = td->runstate == TD_VERIFYING ||
1241 td->io_hist_len || td->verify_batch;
1242 if (td->io_hist_len && td->o.verify_backlog)
1244 td->io_hist_len % td->o.verify_backlog;
1245 if (!verify_data_left)
1246 zbd_reset_zones(td, f, zb, ze);
1249 zbd_reset_write_cnt(td, f);
1252 /* Return random zone index for one of the open zones. */
1253 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1254 const struct io_u *io_u)
1256 return (io_u->offset - f->file_offset) *
1257 f->zbd_info->num_open_zones / f->io_size;
1260 static bool any_io_in_flight(void)
1262 struct thread_data *td;
1265 for_each_td(td, i) {
1266 if (td->io_u_in_flight)
1274 * Modify the offset of an I/O unit that does not refer to an open zone such
1275 * that it refers to an open zone. Close an open zone and open a new zone if
1276 * necessary. The open zone is searched across sequential zones.
1277 * This algorithm can only work correctly if all write pointers are
1278 * a multiple of the fio block size. The caller must neither hold z->mutex
1279 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1281 static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
1284 const uint64_t min_bs = td->o.min_bs[io_u->ddir];
1285 struct fio_file *f = io_u->file;
1286 struct zoned_block_device_info *zbdi = f->zbd_info;
1287 struct fio_zone_info *z;
1288 unsigned int open_zone_idx = -1;
1289 uint32_t zone_idx, new_zone_idx;
1291 bool wait_zone_close;
1293 bool should_retry = true;
1295 assert(is_valid_offset(f, io_u->offset));
1297 if (zbdi->max_open_zones || td->o.job_max_open_zones) {
1299 * This statement accesses zbdi->open_zones[] on purpose
1302 zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)];
1304 zone_idx = zbd_offset_to_zone_idx(f, io_u->offset);
1306 if (zone_idx < f->min_zone)
1307 zone_idx = f->min_zone;
1308 else if (zone_idx >= f->max_zone)
1309 zone_idx = f->max_zone - 1;
1312 "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1313 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1316 * Since z->mutex is the outer lock and zbdi->mutex the inner
1317 * lock it can happen that the state of the zone with index zone_idx
1318 * has changed after 'z' has been assigned and before zbdi->mutex
1319 * has been obtained. Hence the loop.
1324 z = zbd_get_zone(f, zone_idx);
1326 zone_lock(td, f, z);
1328 pthread_mutex_lock(&zbdi->mutex);
1331 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1332 zbdi->max_open_zones == 0 &&
1333 td->o.job_max_open_zones == 0)
1335 if (zbdi->num_open_zones == 0) {
1336 dprint(FD_ZBD, "%s(%s): no zones are open\n",
1337 __func__, f->file_name);
1338 goto open_other_zone;
1343 * List of opened zones is per-device, shared across all
1344 * threads. Start with quasi-random candidate zone. Ignore
1345 * zones which don't belong to thread's offset/size area.
1347 open_zone_idx = pick_random_zone_idx(f, io_u);
1348 assert(!open_zone_idx ||
1349 open_zone_idx < zbdi->num_open_zones);
1350 tmp_idx = open_zone_idx;
1352 for (i = 0; i < zbdi->num_open_zones; i++) {
1355 if (tmp_idx >= zbdi->num_open_zones)
1357 tmpz = zbdi->open_zones[tmp_idx];
1358 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1359 open_zone_idx = tmp_idx;
1360 goto found_candidate_zone;
1366 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1367 __func__, f->file_name);
1369 pthread_mutex_unlock(&zbdi->mutex);
1376 found_candidate_zone:
1377 new_zone_idx = zbdi->open_zones[open_zone_idx];
1378 if (new_zone_idx == zone_idx)
1380 zone_idx = new_zone_idx;
1382 pthread_mutex_unlock(&zbdi->mutex);
1388 /* Both z->mutex and zbdi->mutex are held. */
1391 if (zbd_zone_remainder(z) >= min_bs) {
1392 pthread_mutex_unlock(&zbdi->mutex);
1397 /* Check if number of open zones reaches one of limits. */
1399 zbdi->num_open_zones == f->max_zone - f->min_zone ||
1400 (zbdi->max_open_zones &&
1401 zbdi->num_open_zones == zbdi->max_open_zones) ||
1402 (td->o.job_max_open_zones &&
1403 td->num_open_zones == td->o.job_max_open_zones);
1405 pthread_mutex_unlock(&zbdi->mutex);
1407 /* Only z->mutex is held. */
1410 * When number of open zones reaches to one of limits, wait for
1411 * zone close before opening a new zone.
1413 if (wait_zone_close) {
1415 "%s(%s): quiesce to allow open zones to close\n",
1416 __func__, f->file_name);
1421 /* Zone 'z' is full, so try to open a new zone. */
1422 for (i = f->io_size / zbdi->zone_size; i > 0; i--) {
1427 if (!is_valid_offset(f, z->start)) {
1429 zone_idx = f->min_zone;
1430 z = zbd_get_zone(f, zone_idx);
1432 assert(is_valid_offset(f, z->start));
1435 zone_lock(td, f, z);
1438 if (zbd_open_zone(td, f, z))
1442 /* Only z->mutex is held. */
1444 /* Check whether the write fits in any of the already opened zones. */
1445 pthread_mutex_lock(&zbdi->mutex);
1446 for (i = 0; i < zbdi->num_open_zones; i++) {
1447 zone_idx = zbdi->open_zones[i];
1448 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1450 pthread_mutex_unlock(&zbdi->mutex);
1453 z = zbd_get_zone(f, zone_idx);
1455 zone_lock(td, f, z);
1456 if (zbd_zone_remainder(z) >= min_bs)
1458 pthread_mutex_lock(&zbdi->mutex);
1462 * When any I/O is in-flight or when all I/Os in-flight get completed,
1463 * the I/Os might have closed zones then retry the steps to open a zone.
1464 * Before retry, call io_u_quiesce() to complete in-flight writes.
1466 in_flight = any_io_in_flight();
1467 if (in_flight || should_retry) {
1469 "%s(%s): wait zone close and retry open zones\n",
1470 __func__, f->file_name);
1471 pthread_mutex_unlock(&zbdi->mutex);
1474 zone_lock(td, f, z);
1475 should_retry = in_flight;
1479 pthread_mutex_unlock(&zbdi->mutex);
1483 dprint(FD_ZBD, "%s(%s): did not open another zone\n",
1484 __func__, f->file_name);
1489 dprint(FD_ZBD, "%s(%s): returning zone %d\n",
1490 __func__, f->file_name, zone_idx);
1492 io_u->offset = z->start;
1494 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1500 * Find another zone which has @min_bytes of readable data. Search in zones
1501 * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf.
1503 * Either returns NULL or returns a zone pointer. When the zone has write
1504 * pointer, hold the mutex for the zone.
1506 static struct fio_zone_info *
1507 zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes,
1508 struct fio_zone_info *zb, struct fio_zone_info *zl)
1510 struct fio_file *f = io_u->file;
1511 struct fio_zone_info *z1, *z2;
1512 const struct fio_zone_info *const zf = zbd_get_zone(f, f->min_zone);
1515 * Skip to the next non-empty zone in case of sequential I/O and to
1516 * the nearest non-empty zone in case of random I/O.
1518 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1519 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1521 zone_lock(td, f, z1);
1522 if (z1->start + min_bytes <= z1->wp)
1526 } else if (!td_random(td)) {
1530 if (td_random(td) && z2 >= zf &&
1531 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1533 zone_lock(td, f, z2);
1534 if (z2->start + min_bytes <= z2->wp)
1542 "%s: no zone has %"PRIu64" bytes of readable data\n",
1543 f->file_name, min_bytes);
1549 * zbd_end_zone_io - update zone status at command completion
1551 * @z: zone info pointer
1553 * If the write command made the zone full, close it.
1555 * The caller must hold z->mutex.
1557 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1558 struct fio_zone_info *z)
1560 const struct fio_file *f = io_u->file;
1562 if (io_u->ddir == DDIR_WRITE &&
1563 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1564 pthread_mutex_lock(&f->zbd_info->mutex);
1565 zbd_close_zone(td, f, z);
1566 pthread_mutex_unlock(&f->zbd_info->mutex);
1571 * zbd_queue_io - update the write pointer of a sequential zone
1573 * @success: Whether or not the I/O unit has been queued successfully
1574 * @q: queueing status (busy, completed or queued).
1576 * For write and trim operations, update the write pointer of the I/O unit
1579 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1582 const struct fio_file *f = io_u->file;
1583 struct zoned_block_device_info *zbd_info = f->zbd_info;
1584 struct fio_zone_info *z;
1589 z = zbd_offset_to_zone(f, io_u->offset);
1596 "%s: queued I/O (%lld, %llu) for zone %u\n",
1597 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1599 switch (io_u->ddir) {
1601 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1602 zbd_zone_capacity_end(z));
1605 * z->wp > zone_end means that one or more I/O errors
1608 pthread_mutex_lock(&zbd_info->mutex);
1609 if (z->wp <= zone_end)
1610 zbd_info->wp_valid_data_bytes += zone_end - z->wp;
1611 pthread_mutex_unlock(&zbd_info->mutex);
1618 if (q == FIO_Q_COMPLETED && !io_u->error)
1619 zbd_end_zone_io(td, io_u, z);
1622 if (!success || q != FIO_Q_QUEUED) {
1623 /* BUSY or COMPLETED: unlock the zone */
1625 io_u->zbd_put_io = NULL;
1630 * zbd_put_io - Unlock an I/O unit target zone lock
1633 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1635 const struct fio_file *f = io_u->file;
1636 struct zoned_block_device_info *zbd_info = f->zbd_info;
1637 struct fio_zone_info *z;
1641 z = zbd_offset_to_zone(f, io_u->offset);
1645 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1646 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1648 zbd_end_zone_io(td, io_u, z);
1654 * Windows and MacOS do not define this.
1657 #define EREMOTEIO 121 /* POSIX value */
1660 bool zbd_unaligned_write(int error_code)
1662 switch (error_code) {
1671 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1672 * @td: FIO thread data.
1673 * @io_u: FIO I/O unit.
1675 * For sequential workloads, change the file offset to skip zoneskip bytes when
1676 * no more IO can be performed in the current zone.
1677 * - For read workloads, zoneskip is applied when the io has reached the end of
1678 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1679 * - For write workloads, zoneskip is applied when the zone is full.
1680 * This applies only to read and write operations.
1682 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1684 struct fio_file *f = io_u->file;
1685 enum fio_ddir ddir = io_u->ddir;
1686 struct fio_zone_info *z;
1688 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1689 assert(td->o.zone_size);
1690 assert(f->zbd_info);
1692 z = zbd_offset_to_zone(f, f->last_pos[ddir]);
1695 * When the zone capacity is smaller than the zone size and the I/O is
1696 * sequential write, skip to zone end if the latest position is at the
1697 * zone capacity limit.
1699 if (z->capacity < f->zbd_info->zone_size &&
1700 !td_random(td) && ddir == DDIR_WRITE &&
1701 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1703 "%s: Jump from zone capacity limit to zone end:"
1704 " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n",
1705 f->file_name, f->last_pos[ddir],
1706 zbd_zone_end(z), zbd_zone_idx(f, z), z->capacity);
1707 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1708 f->last_pos[ddir] = zbd_zone_end(z);
1712 * zone_skip is valid only for sequential workloads.
1714 if (td_random(td) || !td->o.zone_skip)
1718 * It is time to switch to a new zone if:
1719 * - zone_bytes == zone_size bytes have already been accessed
1720 * - The last position reached the end of the current zone.
1721 * - For reads with td->o.read_beyond_wp == false, the last position
1722 * reached the zone write pointer.
1724 if (td->zone_bytes >= td->o.zone_size ||
1725 f->last_pos[ddir] >= zbd_zone_end(z) ||
1726 (ddir == DDIR_READ &&
1727 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1732 f->file_offset += td->o.zone_size + td->o.zone_skip;
1735 * Wrap from the beginning, if we exceed the file size
1737 if (f->file_offset >= f->real_file_size)
1738 f->file_offset = get_start_offset(td, f);
1740 f->last_pos[ddir] = f->file_offset;
1741 td->io_skip_bytes += td->o.zone_skip;
1746 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1748 * @td: FIO thread data.
1749 * @io_u: FIO I/O unit.
1750 * @ddir: I/O direction before adjustment.
1752 * Return adjusted I/O direction.
1754 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1758 * In case read direction is chosen for the first random I/O, fio with
1759 * zonemode=zbd stops because no data can be read from zoned block
1760 * devices with all empty zones. Overwrite the first I/O direction as
1761 * write to make sure data to read exists.
1763 assert(io_u->file->zbd_info);
1764 if (ddir != DDIR_READ || !td_rw(td))
1767 if (io_u->file->last_start[DDIR_WRITE] != -1ULL || td->o.read_beyond_wp)
1774 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1775 * @td: FIO thread data.
1776 * @io_u: FIO I/O unit.
1778 * Locking strategy: returns with z->mutex locked if and only if z refers
1779 * to a sequential zone and if io_u_accept is returned. z is the zone that
1780 * corresponds to io_u->offset at the end of this function.
1782 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1784 struct fio_file *f = io_u->file;
1785 struct zoned_block_device_info *zbdi = f->zbd_info;
1786 struct fio_zone_info *zb, *zl, *orig_zb;
1787 uint32_t orig_len = io_u->buflen;
1788 uint64_t min_bs = td->o.min_bs[io_u->ddir];
1794 assert(is_valid_offset(f, io_u->offset));
1795 assert(io_u->buflen);
1797 zb = zbd_offset_to_zone(f, io_u->offset);
1801 /* Accept non-write I/Os for conventional zones. */
1802 if (io_u->ddir != DDIR_WRITE)
1806 * Make sure that writes to conventional zones
1807 * don't cross over to any sequential zones.
1809 if (!(zb + 1)->has_wp ||
1810 io_u->offset + io_u->buflen <= (zb + 1)->start)
1813 if (io_u->offset + min_bs > (zb + 1)->start) {
1815 "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n",
1816 f->file_name, io_u->offset,
1817 min_bs, (zb + 1)->start);
1819 zb->start + (zb + 1)->start - io_u->offset;
1820 new_len = min(io_u->buflen,
1821 (zb + 1)->start - io_u->offset);
1823 new_len = (zb + 1)->start - io_u->offset;
1826 io_u->buflen = new_len / min_bs * min_bs;
1832 * Accept the I/O offset for reads if reading beyond the write pointer
1835 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1836 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1839 zone_lock(td, f, zb);
1841 switch (io_u->ddir) {
1843 if (td->runstate == TD_VERIFYING && td_write(td))
1847 * Check that there is enough written data in the zone to do an
1848 * I/O of at least min_bs B. If there isn't, find a new zone for
1851 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1852 zb->wp - zb->start : 0;
1853 if (range < min_bs ||
1854 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1856 zl = zbd_get_zone(f, f->max_zone);
1857 zb = zbd_find_zone(td, io_u, min_bs, zb, zl);
1860 "%s: zbd_find_zone(%lld, %llu) failed\n",
1861 f->file_name, io_u->offset,
1866 * zbd_find_zone() returned a zone with a range of at
1869 range = zb->wp - zb->start;
1870 assert(range >= min_bs);
1873 io_u->offset = zb->start;
1877 * Make sure the I/O is within the zone valid data range while
1878 * maximizing the I/O size and preserving randomness.
1880 if (range <= io_u->buflen)
1881 io_u->offset = zb->start;
1882 else if (td_random(td))
1883 io_u->offset = zb->start +
1884 ((io_u->offset - orig_zb->start) %
1885 (range - io_u->buflen)) / min_bs * min_bs;
1888 * When zbd_find_zone() returns a conventional zone,
1889 * we can simply accept the new i/o offset here.
1895 * Make sure the I/O does not cross over the zone wp position.
1897 new_len = min((unsigned long long)io_u->buflen,
1898 (unsigned long long)(zb->wp - io_u->offset));
1899 new_len = new_len / min_bs * min_bs;
1900 if (new_len < io_u->buflen) {
1901 io_u->buflen = new_len;
1902 dprint(FD_IO, "Changed length from %u into %llu\n",
1903 orig_len, io_u->buflen);
1906 assert(zb->start <= io_u->offset);
1907 assert(io_u->offset + io_u->buflen <= zb->wp);
1912 if (io_u->buflen > zbdi->zone_size) {
1913 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
1915 "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n",
1916 f->file_name, io_u->buflen, zbdi->zone_size);
1921 if (zbd_zone_remainder(zb) > 0 &&
1922 zbd_zone_remainder(zb) < min_bs) {
1923 pthread_mutex_lock(&f->zbd_info->mutex);
1924 zbd_close_zone(td, f, zb);
1925 pthread_mutex_unlock(&f->zbd_info->mutex);
1927 "%s: finish zone %d\n",
1928 f->file_name, zbd_zone_idx(f, zb));
1930 zbd_finish_zone(td, f, zb);
1931 if (zbd_zone_idx(f, zb) + 1 >= f->max_zone) {
1937 /* Find the next write pointer zone */
1940 if (zbd_zone_idx(f, zb) >= f->max_zone)
1941 zb = zbd_get_zone(f, f->min_zone);
1942 } while (!zb->has_wp);
1944 zone_lock(td, f, zb);
1947 if (!zbd_open_zone(td, f, zb)) {
1949 zb = zbd_convert_to_open_zone(td, io_u);
1951 dprint(FD_IO, "%s: can't convert to open zone",
1957 if (zbd_zone_remainder(zb) > 0 &&
1958 zbd_zone_remainder(zb) < min_bs)
1961 /* Check whether the zone reset threshold has been exceeded */
1962 if (td->o.zrf.u.f) {
1963 if (zbdi->wp_valid_data_bytes >=
1964 f->io_size * td->o.zrt.u.f &&
1965 zbd_dec_and_reset_write_cnt(td, f))
1969 /* Reset the zone pointer if necessary */
1970 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
1971 if (td->o.verify != VERIFY_NONE) {
1973 * Unset io-u->file to tell get_next_verify()
1974 * that this IO is not requeue.
1977 if (!get_next_verify(td, io_u)) {
1985 * Since previous write requests may have been submitted
1986 * asynchronously and since we will submit the zone
1987 * reset synchronously, wait until previously submitted
1988 * write requests have completed before issuing a
1993 if (zbd_reset_zone(td, f, zb) < 0)
1996 if (zb->capacity < min_bs) {
1997 td_verror(td, EINVAL, "ZCAP is less min_bs");
1998 log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n",
1999 zb->capacity, min_bs);
2004 /* Make writes occur at the write pointer */
2005 assert(!zbd_zone_full(f, zb, min_bs));
2006 io_u->offset = zb->wp;
2007 if (!is_valid_offset(f, io_u->offset)) {
2008 td_verror(td, EINVAL, "invalid WP value");
2009 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
2010 f->file_name, io_u->offset);
2015 * Make sure that the buflen is a multiple of the minimal
2016 * block size. Give up if shrinking would make the request too
2019 new_len = min((unsigned long long)io_u->buflen,
2020 zbd_zone_capacity_end(zb) - io_u->offset);
2021 new_len = new_len / min_bs * min_bs;
2022 if (new_len == io_u->buflen)
2024 if (new_len >= min_bs) {
2025 io_u->buflen = new_len;
2026 dprint(FD_IO, "Changed length from %u into %llu\n",
2027 orig_len, io_u->buflen);
2031 td_verror(td, EIO, "zone remainder too small");
2032 log_err("zone remainder %lld smaller than min block size %"PRIu64"\n",
2033 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
2038 /* Check random trim targets a non-empty zone */
2039 if (!td_random(td) || zb->wp > zb->start)
2042 /* Find out a non-empty zone to trim */
2044 zl = zbd_get_zone(f, f->max_zone);
2045 zb = zbd_find_zone(td, io_u, 1, zb, zl);
2047 io_u->offset = zb->start;
2048 dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n",
2049 f->file_name, io_u->offset);
2058 case DDIR_SYNC_FILE_RANGE:
2069 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
2070 assert(!io_u->zbd_queue_io);
2071 assert(!io_u->zbd_put_io);
2073 io_u->zbd_queue_io = zbd_queue_io;
2074 io_u->zbd_put_io = zbd_put_io;
2077 * Since we return with the zone lock still held,
2078 * add an annotation to let Coverity know that it
2081 /* coverity[missing_unlock] */
2086 if (zb && zb->has_wp)
2092 /* Return a string with ZBD statistics */
2093 char *zbd_write_status(const struct thread_stat *ts)
2097 if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0)
2103 * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim
2105 * @td: FIO thread data.
2106 * @io_u: FIO I/O unit.
2108 * It is assumed that z->mutex is already locked.
2109 * Return io_u_completed when reset zone succeeds. Return 0 when the target zone
2110 * does not have write pointer. On error, return negative errno.
2112 int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u)
2114 struct fio_file *f = io_u->file;
2115 struct fio_zone_info *z;
2118 z = zbd_offset_to_zone(f, io_u->offset);
2122 if (io_u->offset != z->start) {
2123 log_err("Trim offset not at zone start (%lld)\n",
2128 ret = zbd_reset_zone((struct thread_data *)td, f, z);
2132 return io_u_completed;