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_nr(const struct fio_file *f,
31 struct fio_zone_info *zone)
33 return zone - f->zbd_info->zone_info;
37 * zbd_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 uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset)
46 if (f->zbd_info->zone_size_log2 > 0)
47 zone_idx = offset >> f->zbd_info->zone_size_log2;
49 zone_idx = offset / f->zbd_info->zone_size;
51 return min(zone_idx, f->zbd_info->nr_zones);
55 * zbd_zone_end - Return zone end location
56 * @z: zone info pointer.
58 static inline uint64_t zbd_zone_end(const struct fio_zone_info *z)
64 * zbd_zone_capacity_end - Return zone capacity limit end location
65 * @z: zone info pointer.
67 static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z)
69 return z->start + z->capacity;
73 * zbd_zone_full - verify whether a minimum number of bytes remain in a zone
75 * @z: zone info pointer.
76 * @required: minimum number of bytes that must remain in a zone.
78 * The caller must hold z->mutex.
80 static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z,
83 assert((required & 511) == 0);
86 z->wp + required > zbd_zone_capacity_end(z);
89 static void zone_lock(struct thread_data *td, const struct fio_file *f,
90 struct fio_zone_info *z)
92 struct zoned_block_device_info *zbd = f->zbd_info;
93 uint32_t nz = z - zbd->zone_info;
95 /* A thread should never lock zones outside its working area. */
96 assert(f->min_zone <= nz && nz < f->max_zone);
101 * Lock the io_u target zone. The zone will be unlocked if io_u offset
102 * is changed or when io_u completes and zbd_put_io() executed.
103 * To avoid multiple jobs doing asynchronous I/Os from deadlocking each
104 * other waiting for zone locks when building an io_u batch, first
105 * only trylock the zone. If the zone is already locked by another job,
106 * process the currently queued I/Os so that I/O progress is made and
109 if (pthread_mutex_trylock(&z->mutex) != 0) {
110 if (!td_ioengine_flagged(td, FIO_SYNCIO))
112 pthread_mutex_lock(&z->mutex);
116 static inline void zone_unlock(struct fio_zone_info *z)
121 ret = pthread_mutex_unlock(&z->mutex);
125 static inline struct fio_zone_info *get_zone(const struct fio_file *f,
126 unsigned int zone_nr)
128 return &f->zbd_info->zone_info[zone_nr];
132 * zbd_get_zoned_model - Get a device zoned model
133 * @td: FIO thread data
134 * @f: FIO file for which to get model information
136 static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f,
137 enum zbd_zoned_model *model)
141 if (f->filetype == FIO_TYPE_PIPE) {
142 log_err("zonemode=zbd does not support pipes\n");
146 /* If regular file, always emulate zones inside the file. */
147 if (f->filetype == FIO_TYPE_FILE) {
152 if (td->io_ops && td->io_ops->get_zoned_model)
153 ret = td->io_ops->get_zoned_model(td, f, model);
155 ret = blkzoned_get_zoned_model(td, f, model);
157 td_verror(td, errno, "get zoned model failed");
158 log_err("%s: get zoned model failed (%d).\n",
159 f->file_name, errno);
166 * zbd_report_zones - Get zone information
167 * @td: FIO thread data.
168 * @f: FIO file for which to get zone information
169 * @offset: offset from which to report zones
170 * @zones: Array of struct zbd_zone
171 * @nr_zones: Size of @zones array
173 * Get zone information into @zones starting from the zone at offset @offset
174 * for the device specified by @f.
176 * Returns the number of zones reported upon success and a negative error code
177 * upon failure. If the zone report is empty, always assume an error (device
178 * problem) and return -EIO.
180 static int zbd_report_zones(struct thread_data *td, struct fio_file *f,
181 uint64_t offset, struct zbd_zone *zones,
182 unsigned int nr_zones)
186 if (td->io_ops && td->io_ops->report_zones)
187 ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones);
189 ret = blkzoned_report_zones(td, f, offset, zones, nr_zones);
191 td_verror(td, errno, "report zones failed");
192 log_err("%s: report zones from sector %"PRIu64" failed (%d).\n",
193 f->file_name, offset >> 9, errno);
194 } else if (ret == 0) {
195 td_verror(td, errno, "Empty zone report");
196 log_err("%s: report zones from sector %"PRIu64" is empty.\n",
197 f->file_name, offset >> 9);
205 * zbd_reset_wp - reset the write pointer of a range of zones
206 * @td: FIO thread data.
207 * @f: FIO file for which to reset zones
208 * @offset: Starting offset of the first zone to reset
209 * @length: Length of the range of zones to reset
211 * Reset the write pointer of all zones in the range @offset...@offset+@length.
212 * Returns 0 upon success and a negative error code upon failure.
214 static int zbd_reset_wp(struct thread_data *td, struct fio_file *f,
215 uint64_t offset, uint64_t length)
219 if (td->io_ops && td->io_ops->reset_wp)
220 ret = td->io_ops->reset_wp(td, f, offset, length);
222 ret = blkzoned_reset_wp(td, f, offset, length);
224 td_verror(td, errno, "resetting wp failed");
225 log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n",
226 f->file_name, length >> 9, offset >> 9, errno);
233 * zbd_reset_zone - reset the write pointer of a single zone
234 * @td: FIO thread data.
235 * @f: FIO file associated with the disk for which to reset a write pointer.
238 * Returns 0 upon success and a negative error code upon failure.
240 * The caller must hold z->mutex.
242 static int zbd_reset_zone(struct thread_data *td, struct fio_file *f,
243 struct fio_zone_info *z)
245 uint64_t offset = z->start;
246 uint64_t length = (z+1)->start - offset;
247 uint64_t data_in_zone = z->wp - z->start;
253 assert(is_valid_offset(f, offset + length - 1));
255 dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", f->file_name,
257 switch (f->zbd_info->model) {
259 case ZBD_HOST_MANAGED:
260 ret = zbd_reset_wp(td, f, offset, length);
268 pthread_mutex_lock(&f->zbd_info->mutex);
269 f->zbd_info->sectors_with_data -= data_in_zone;
270 f->zbd_info->wp_sectors_with_data -= data_in_zone;
271 pthread_mutex_unlock(&f->zbd_info->mutex);
275 td->ts.nr_zone_resets++;
281 * zbd_close_zone - Remove a zone from the open zones array.
282 * @td: FIO thread data.
283 * @f: FIO file associated with the disk for which to reset a write pointer.
284 * @zone_idx: Index of the zone to remove.
286 * The caller must hold f->zbd_info->mutex.
288 static void zbd_close_zone(struct thread_data *td, const struct fio_file *f,
289 unsigned int zone_idx)
291 uint32_t open_zone_idx = 0;
293 for (; open_zone_idx < f->zbd_info->num_open_zones; open_zone_idx++) {
294 if (f->zbd_info->open_zones[open_zone_idx] == zone_idx)
297 if (open_zone_idx == f->zbd_info->num_open_zones)
300 dprint(FD_ZBD, "%s: closing zone %d\n", f->file_name, zone_idx);
301 memmove(f->zbd_info->open_zones + open_zone_idx,
302 f->zbd_info->open_zones + open_zone_idx + 1,
303 (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) *
304 sizeof(f->zbd_info->open_zones[0]));
305 f->zbd_info->num_open_zones--;
306 td->num_open_zones--;
307 get_zone(f, zone_idx)->open = 0;
311 * zbd_reset_zones - Reset a range of zones.
312 * @td: fio thread data.
313 * @f: fio file for which to reset zones
314 * @zb: first zone to reset.
315 * @ze: first zone not to reset.
317 * Returns 0 upon success and 1 upon failure.
319 static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
320 struct fio_zone_info *const zb,
321 struct fio_zone_info *const ze)
323 struct fio_zone_info *z;
324 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
329 dprint(FD_ZBD, "%s: examining zones %u .. %u\n", f->file_name,
330 zbd_zone_nr(f, zb), zbd_zone_nr(f, ze));
331 for (z = zb; z < ze; z++) {
332 uint32_t nz = zbd_zone_nr(f, z);
337 pthread_mutex_lock(&f->zbd_info->mutex);
338 zbd_close_zone(td, f, nz);
339 pthread_mutex_unlock(&f->zbd_info->mutex);
340 if (z->wp != z->start) {
341 dprint(FD_ZBD, "%s: resetting zone %u\n",
342 f->file_name, zbd_zone_nr(f, z));
343 if (zbd_reset_zone(td, f, z) < 0)
353 * zbd_get_max_open_zones - Get the maximum number of open zones
354 * @td: FIO thread data
355 * @f: FIO file for which to get max open zones
356 * @max_open_zones: Upon success, result will be stored here.
358 * A @max_open_zones value set to zero means no limit.
360 * Returns 0 upon success and a negative error code upon failure.
362 static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f,
363 unsigned int *max_open_zones)
367 if (td->io_ops && td->io_ops->get_max_open_zones)
368 ret = td->io_ops->get_max_open_zones(td, f, max_open_zones);
370 ret = blkzoned_get_max_open_zones(td, f, max_open_zones);
372 td_verror(td, errno, "get max open zones failed");
373 log_err("%s: get max open zones failed (%d).\n",
374 f->file_name, errno);
381 * is_zone_open - Test if a zone is already in the array of open zones.
382 * @td: fio thread data.
383 * @f: fio file for which to test zones.
384 * @zone_idx: Index of the zone to check.
386 * The caller must hold f->zbd_info->mutex.
388 static bool is_zone_open(const struct thread_data *td, const struct fio_file *f,
389 unsigned int zone_idx)
391 struct zoned_block_device_info *zbdi = f->zbd_info;
395 * This function should never be called when zbdi->max_open_zones == 0.
397 assert(zbdi->max_open_zones);
398 assert(td->o.job_max_open_zones == 0 ||
399 td->num_open_zones <= td->o.job_max_open_zones);
400 assert(td->o.job_max_open_zones <= zbdi->max_open_zones);
401 assert(zbdi->num_open_zones <= zbdi->max_open_zones);
403 for (i = 0; i < zbdi->num_open_zones; i++)
404 if (zbdi->open_zones[i] == zone_idx)
411 * zbd_open_zone - Add a zone to the array of open zones.
412 * @td: fio thread data.
413 * @f: fio file that has the open zones to add.
414 * @zone_idx: Index of the zone to add.
416 * Open a ZBD zone if it is not already open. Returns true if either the zone
417 * was already open or if the zone was successfully added to the array of open
418 * zones without exceeding the maximum number of open zones. Returns false if
419 * the zone was not already open and opening the zone would cause the zone limit
422 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
425 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
426 struct zoned_block_device_info *zbdi = f->zbd_info;
427 struct fio_zone_info *z = get_zone(f, zone_idx);
430 if (z->cond == ZBD_ZONE_COND_OFFLINE)
434 * Skip full zones with data verification enabled because resetting a
435 * zone causes data loss and hence causes verification to fail.
437 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
441 * zbdi->max_open_zones == 0 means that there is no limit on the maximum
442 * number of open zones. In this case, do no track open zones in
443 * zbdi->open_zones array.
445 if (!zbdi->max_open_zones)
448 pthread_mutex_lock(&zbdi->mutex);
449 if (is_zone_open(td, f, zone_idx)) {
451 * If the zone is already open and going to be full by writes
452 * in-flight, handle it as a full zone instead of an open zone.
454 if (z->wp >= zbd_zone_capacity_end(z))
459 /* Zero means no limit */
460 if (td->o.job_max_open_zones > 0 &&
461 td->num_open_zones >= td->o.job_max_open_zones)
463 if (zbdi->num_open_zones >= zbdi->max_open_zones)
465 dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx);
466 zbdi->open_zones[zbdi->num_open_zones++] = zone_idx;
467 td->num_open_zones++;
472 pthread_mutex_unlock(&zbdi->mutex);
476 /* Verify whether direct I/O is used for all host-managed zoned drives. */
477 static bool zbd_using_direct_io(void)
479 struct thread_data *td;
484 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
486 for_each_file(td, f, j) {
488 f->zbd_info->model == ZBD_HOST_MANAGED)
496 /* Whether or not the I/O range for f includes one or more sequential zones */
497 static bool zbd_is_seq_job(struct fio_file *f)
499 uint32_t zone_idx, zone_idx_b, zone_idx_e;
504 zone_idx_b = zbd_zone_idx(f, f->file_offset);
505 zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1);
506 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
507 if (get_zone(f, zone_idx)->has_wp)
514 * Verify whether offset and size parameters are aligned with zone boundaries.
516 static bool zbd_verify_sizes(void)
518 const struct fio_zone_info *z;
519 struct thread_data *td;
521 uint64_t new_offset, new_end;
526 for_each_file(td, f, j) {
529 if (f->file_offset >= f->real_file_size)
531 if (!zbd_is_seq_job(f))
534 if (!td->o.zone_size) {
535 td->o.zone_size = f->zbd_info->zone_size;
536 if (!td->o.zone_size) {
537 log_err("%s: invalid 0 zone size\n",
541 } else if (td->o.zone_size != f->zbd_info->zone_size) {
542 log_err("%s: job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
543 f->file_name, td->o.zone_size,
544 f->zbd_info->zone_size);
548 if (td->o.zone_skip % td->o.zone_size) {
549 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
550 f->file_name, td->o.zone_skip,
555 zone_idx = zbd_zone_idx(f, f->file_offset);
556 z = get_zone(f, zone_idx);
557 if ((f->file_offset != z->start) &&
558 (td->o.td_ddir != TD_DDIR_READ)) {
559 new_offset = zbd_zone_end(z);
560 if (new_offset >= f->file_offset + f->io_size) {
561 log_info("%s: io_size must be at least one zone\n",
565 log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n",
566 f->file_name, f->file_offset,
568 f->io_size -= (new_offset - f->file_offset);
569 f->file_offset = new_offset;
571 zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size);
572 z = get_zone(f, zone_idx);
574 if ((td->o.td_ddir != TD_DDIR_READ) &&
575 (f->file_offset + f->io_size != new_end)) {
576 if (new_end <= f->file_offset) {
577 log_info("%s: io_size must be at least one zone\n",
581 log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n",
582 f->file_name, f->io_size,
583 new_end - f->file_offset);
584 f->io_size = new_end - f->file_offset;
592 static bool zbd_verify_bs(void)
594 struct thread_data *td;
600 (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] ||
601 td->o.bssplit_nr[DDIR_TRIM])) {
602 log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n");
605 for_each_file(td, f, j) {
610 zone_size = f->zbd_info->zone_size;
611 if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) {
612 log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n",
613 f->file_name, td->o.bs[DDIR_TRIM],
617 for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) {
618 if (td->o.verify != VERIFY_NONE &&
619 zone_size % td->o.bs[k] != 0) {
620 log_info("%s: block size %llu is not a divisor of the zone size %"PRIu64"\n",
621 f->file_name, td->o.bs[k],
631 static int ilog2(uint64_t i)
643 * Initialize f->zbd_info for devices that are not zoned block devices. This
644 * allows to execute a ZBD workload against a non-ZBD device.
646 static int init_zone_info(struct thread_data *td, struct fio_file *f)
649 struct fio_zone_info *p;
650 uint64_t zone_size = td->o.zone_size;
651 uint64_t zone_capacity = td->o.zone_capacity;
652 struct zoned_block_device_info *zbd_info = NULL;
655 if (zone_size == 0) {
656 log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n",
661 if (zone_size < 512) {
662 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
667 if (zone_capacity == 0)
668 zone_capacity = zone_size;
670 if (zone_capacity > zone_size) {
671 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
672 f->file_name, td->o.zone_capacity, td->o.zone_size);
676 if (f->real_file_size < zone_size) {
677 log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n",
678 f->file_name, f->real_file_size, zone_size);
682 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
683 zbd_info = scalloc(1, sizeof(*zbd_info) +
684 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
688 mutex_init_pshared(&zbd_info->mutex);
689 zbd_info->refcount = 1;
690 p = &zbd_info->zone_info[0];
691 for (i = 0; i < nr_zones; i++, p++) {
692 mutex_init_pshared_with_type(&p->mutex,
693 PTHREAD_MUTEX_RECURSIVE);
694 p->start = i * zone_size;
696 p->type = ZBD_ZONE_TYPE_SWR;
697 p->cond = ZBD_ZONE_COND_EMPTY;
698 p->capacity = zone_capacity;
702 p->start = nr_zones * zone_size;
704 f->zbd_info = zbd_info;
705 f->zbd_info->zone_size = zone_size;
706 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
707 ilog2(zone_size) : 0;
708 f->zbd_info->nr_zones = nr_zones;
713 * Maximum number of zones to report in one operation.
715 #define ZBD_REPORT_MAX_ZONES 8192U
718 * Parse the device zone report and store it in f->zbd_info. Must be called
719 * only for devices that are zoned, namely those with a model != ZBD_NONE.
721 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
724 struct zbd_zone *zones, *z;
725 struct fio_zone_info *p;
726 uint64_t zone_size, offset;
727 struct zoned_block_device_info *zbd_info = NULL;
728 int i, j, ret = -ENOMEM;
730 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
734 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
737 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
742 zone_size = zones[0].len;
743 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
745 if (td->o.zone_size == 0) {
746 td->o.zone_size = zone_size;
747 } else if (td->o.zone_size != zone_size) {
748 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
749 f->file_name, td->o.zone_size, zone_size);
754 dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n", f->file_name,
755 nr_zones, zone_size / 1024);
757 zbd_info = scalloc(1, sizeof(*zbd_info) +
758 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
761 mutex_init_pshared(&zbd_info->mutex);
762 zbd_info->refcount = 1;
763 p = &zbd_info->zone_info[0];
764 for (offset = 0, j = 0; j < nr_zones;) {
766 for (i = 0; i < nrz; i++, j++, z++, p++) {
767 mutex_init_pshared_with_type(&p->mutex,
768 PTHREAD_MUTEX_RECURSIVE);
770 p->capacity = z->capacity;
772 case ZBD_ZONE_COND_NOT_WP:
773 case ZBD_ZONE_COND_FULL:
774 p->wp = p->start + p->capacity;
777 assert(z->start <= z->wp);
778 assert(z->wp <= z->start + zone_size);
784 case ZBD_ZONE_TYPE_SWR:
793 if (j > 0 && p->start != p[-1].start + zone_size) {
794 log_info("%s: invalid zone data\n",
801 offset = z->start + z->len;
804 nrz = zbd_report_zones(td, f, offset, zones,
805 min((uint32_t)(nr_zones - j),
806 ZBD_REPORT_MAX_ZONES));
809 log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n",
810 offset, f->file_name, -ret);
816 zbd_info->zone_info[nr_zones].start = offset;
818 f->zbd_info = zbd_info;
819 f->zbd_info->zone_size = zone_size;
820 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
821 ilog2(zone_size) : 0;
822 f->zbd_info->nr_zones = nr_zones;
832 static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f)
834 struct zoned_block_device_info *zbd = f->zbd_info;
835 unsigned int max_open_zones;
838 if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) {
839 /* Only host-managed devices have a max open limit */
840 zbd->max_open_zones = td->o.max_open_zones;
844 /* If host-managed, get the max open limit */
845 ret = zbd_get_max_open_zones(td, f, &max_open_zones);
849 if (!max_open_zones) {
850 /* No device limit */
851 zbd->max_open_zones = td->o.max_open_zones;
852 } else if (!td->o.max_open_zones) {
853 /* No user limit. Set limit to device limit */
854 zbd->max_open_zones = max_open_zones;
855 } else if (td->o.max_open_zones <= max_open_zones) {
856 /* Both user limit and dev limit. User limit not too large */
857 zbd->max_open_zones = td->o.max_open_zones;
859 /* Both user limit and dev limit. User limit too large */
860 td_verror(td, EINVAL,
861 "Specified --max_open_zones is too large");
862 log_err("Specified --max_open_zones (%d) is larger than max (%u)\n",
863 td->o.max_open_zones, max_open_zones);
868 /* Ensure that the limit is not larger than FIO's internal limit */
869 if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) {
870 td_verror(td, EINVAL, "'max_open_zones' value is too large");
871 log_err("'max_open_zones' value is larger than %u\n", ZBD_MAX_OPEN_ZONES);
875 dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n",
876 f->file_name, zbd->max_open_zones);
882 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
884 * Returns 0 upon success and a negative error code upon failure.
886 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
888 enum zbd_zoned_model zbd_model;
891 assert(td->o.zone_mode == ZONE_MODE_ZBD);
893 ret = zbd_get_zoned_model(td, f, &zbd_model);
899 case ZBD_HOST_MANAGED:
900 ret = parse_zone_info(td, f);
905 ret = init_zone_info(td, f);
910 td_verror(td, EINVAL, "Unsupported zoned model");
911 log_err("Unsupported zoned model\n");
916 f->zbd_info->model = zbd_model;
918 ret = zbd_set_max_open_zones(td, f);
920 zbd_free_zone_info(f);
927 void zbd_free_zone_info(struct fio_file *f)
933 pthread_mutex_lock(&f->zbd_info->mutex);
934 refcount = --f->zbd_info->refcount;
935 pthread_mutex_unlock(&f->zbd_info->mutex);
937 assert((int32_t)refcount >= 0);
944 * Initialize f->zbd_info.
946 * Returns 0 upon success and a negative error code upon failure.
948 * Note: this function can only work correctly if it is called before the first
951 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
953 struct thread_data *td2;
957 for_each_td(td2, i) {
958 for_each_file(td2, f2, j) {
959 if (td2 == td && f2 == file)
962 strcmp(f2->file_name, file->file_name) != 0)
964 file->zbd_info = f2->zbd_info;
965 file->zbd_info->refcount++;
970 ret = zbd_create_zone_info(td, file);
972 td_verror(td, -ret, "zbd_create_zone_info() failed");
976 int zbd_init_files(struct thread_data *td)
981 for_each_file(td, f, i) {
982 if (zbd_init_zone_info(td, f))
988 void zbd_recalc_options_with_zone_granularity(struct thread_data *td)
993 for_each_file(td, f, i) {
994 struct zoned_block_device_info *zbd = f->zbd_info;
995 // zonemode=strided doesn't get per-file zone size.
996 uint64_t zone_size = zbd ? zbd->zone_size : td->o.zone_size;
1001 if (td->o.size_nz > 0) {
1002 td->o.size = td->o.size_nz * zone_size;
1004 if (td->o.io_size_nz > 0) {
1005 td->o.io_size = td->o.io_size_nz * zone_size;
1007 if (td->o.start_offset_nz > 0) {
1008 td->o.start_offset = td->o.start_offset_nz * zone_size;
1010 if (td->o.offset_increment_nz > 0) {
1011 td->o.offset_increment = td->o.offset_increment_nz * zone_size;
1013 if (td->o.zone_skip_nz > 0) {
1014 td->o.zone_skip = td->o.zone_skip_nz * zone_size;
1019 int zbd_setup_files(struct thread_data *td)
1024 if (!zbd_using_direct_io()) {
1025 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
1029 if (!zbd_verify_sizes())
1032 if (!zbd_verify_bs())
1035 for_each_file(td, f, i) {
1036 struct zoned_block_device_info *zbd = f->zbd_info;
1037 struct fio_zone_info *z;
1042 f->min_zone = zbd_zone_idx(f, f->file_offset);
1043 f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size);
1046 * When all zones in the I/O range are conventional, io_size
1047 * can be smaller than zone size, making min_zone the same
1048 * as max_zone. This is why the assert below needs to be made
1051 if (zbd_is_seq_job(f))
1052 assert(f->min_zone < f->max_zone);
1054 if (td->o.max_open_zones > 0 &&
1055 zbd->max_open_zones != td->o.max_open_zones) {
1056 log_err("Different 'max_open_zones' values\n");
1061 * The per job max open zones limit cannot be used without a
1062 * global max open zones limit. (As the tracking of open zones
1063 * is disabled when there is no global max open zones limit.)
1065 if (td->o.job_max_open_zones && !zbd->max_open_zones) {
1066 log_err("'job_max_open_zones' cannot be used without a global open zones limit\n");
1071 * zbd->max_open_zones is the global limit shared for all jobs
1072 * that target the same zoned block device. Force sync the per
1073 * thread global limit with the actual global limit. (The real
1074 * per thread/job limit is stored in td->o.job_max_open_zones).
1076 td->o.max_open_zones = zbd->max_open_zones;
1078 for (zi = f->min_zone; zi < f->max_zone; zi++) {
1079 z = &zbd->zone_info[zi];
1080 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
1081 z->cond != ZBD_ZONE_COND_EXP_OPEN)
1083 if (zbd_open_zone(td, f, zi))
1086 * If the number of open zones exceeds specified limits,
1087 * reset all extra open zones.
1089 if (zbd_reset_zone(td, f, z) < 0) {
1090 log_err("Failed to reest zone %d\n", zi);
1100 * Reset zbd_info.write_cnt, the counter that counts down towards the next
1103 static void _zbd_reset_write_cnt(const struct thread_data *td,
1104 const struct fio_file *f)
1106 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
1108 f->zbd_info->write_cnt = td->o.zrf.u.f ?
1109 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
1112 static void zbd_reset_write_cnt(const struct thread_data *td,
1113 const struct fio_file *f)
1115 pthread_mutex_lock(&f->zbd_info->mutex);
1116 _zbd_reset_write_cnt(td, f);
1117 pthread_mutex_unlock(&f->zbd_info->mutex);
1120 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
1121 const struct fio_file *f)
1123 uint32_t write_cnt = 0;
1125 pthread_mutex_lock(&f->zbd_info->mutex);
1126 assert(f->zbd_info->write_cnt);
1127 if (f->zbd_info->write_cnt)
1128 write_cnt = --f->zbd_info->write_cnt;
1130 _zbd_reset_write_cnt(td, f);
1131 pthread_mutex_unlock(&f->zbd_info->mutex);
1133 return write_cnt == 0;
1141 /* Calculate the number of sectors with data (swd) and perform action 'a' */
1142 static uint64_t zbd_process_swd(struct thread_data *td,
1143 const struct fio_file *f, enum swd_action a)
1145 struct fio_zone_info *zb, *ze, *z;
1147 uint64_t wp_swd = 0;
1149 zb = get_zone(f, f->min_zone);
1150 ze = get_zone(f, f->max_zone);
1151 for (z = zb; z < ze; z++) {
1153 zone_lock(td, f, z);
1154 wp_swd += z->wp - z->start;
1156 swd += z->wp - z->start;
1158 pthread_mutex_lock(&f->zbd_info->mutex);
1161 assert(f->zbd_info->sectors_with_data == swd);
1162 assert(f->zbd_info->wp_sectors_with_data == wp_swd);
1165 f->zbd_info->sectors_with_data = swd;
1166 f->zbd_info->wp_sectors_with_data = wp_swd;
1169 pthread_mutex_unlock(&f->zbd_info->mutex);
1170 for (z = zb; z < ze; z++)
1178 * The swd check is useful for debugging but takes too much time to leave
1179 * it enabled all the time. Hence it is disabled by default.
1181 static const bool enable_check_swd = false;
1183 /* Check whether the values of zbd_info.*sectors_with_data are correct. */
1184 static void zbd_check_swd(struct thread_data *td, const struct fio_file *f)
1186 if (!enable_check_swd)
1189 zbd_process_swd(td, f, CHECK_SWD);
1192 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
1194 struct fio_zone_info *zb, *ze;
1197 if (!f->zbd_info || !td_write(td))
1200 zb = get_zone(f, f->min_zone);
1201 ze = get_zone(f, f->max_zone);
1202 swd = zbd_process_swd(td, f, SET_SWD);
1203 dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name,
1206 * If data verification is enabled reset the affected zones before
1207 * writing any data to avoid that a zone reset has to be issued while
1208 * writing data, which causes data loss.
1210 if (td->o.verify != VERIFY_NONE && td->runstate != TD_VERIFYING)
1211 zbd_reset_zones(td, f, zb, ze);
1212 zbd_reset_write_cnt(td, f);
1215 /* Return random zone index for one of the open zones. */
1216 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1217 const struct io_u *io_u)
1219 return (io_u->offset - f->file_offset) * f->zbd_info->num_open_zones /
1223 static bool any_io_in_flight(void)
1225 struct thread_data *td;
1228 for_each_td(td, i) {
1229 if (td->io_u_in_flight)
1237 * Modify the offset of an I/O unit that does not refer to an open zone such
1238 * that it refers to an open zone. Close an open zone and open a new zone if
1239 * necessary. The open zone is searched across sequential zones.
1240 * This algorithm can only work correctly if all write pointers are
1241 * a multiple of the fio block size. The caller must neither hold z->mutex
1242 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1244 static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
1247 const uint64_t min_bs = td->o.min_bs[io_u->ddir];
1248 struct fio_file *f = io_u->file;
1249 struct zoned_block_device_info *zbdi = f->zbd_info;
1250 struct fio_zone_info *z;
1251 unsigned int open_zone_idx = -1;
1252 uint32_t zone_idx, new_zone_idx;
1254 bool wait_zone_close;
1256 bool should_retry = true;
1258 assert(is_valid_offset(f, io_u->offset));
1260 if (zbdi->max_open_zones || td->o.job_max_open_zones) {
1262 * This statement accesses zbdi->open_zones[] on purpose
1265 zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)];
1267 zone_idx = zbd_zone_idx(f, io_u->offset);
1269 if (zone_idx < f->min_zone)
1270 zone_idx = f->min_zone;
1271 else if (zone_idx >= f->max_zone)
1272 zone_idx = f->max_zone - 1;
1273 dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1274 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1277 * Since z->mutex is the outer lock and zbdi->mutex the inner
1278 * lock it can happen that the state of the zone with index zone_idx
1279 * has changed after 'z' has been assigned and before zbdi->mutex
1280 * has been obtained. Hence the loop.
1285 z = get_zone(f, zone_idx);
1287 zone_lock(td, f, z);
1288 pthread_mutex_lock(&zbdi->mutex);
1290 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1291 zbdi->max_open_zones == 0 && td->o.job_max_open_zones == 0)
1293 if (zbdi->num_open_zones == 0) {
1294 dprint(FD_ZBD, "%s(%s): no zones are open\n",
1295 __func__, f->file_name);
1296 goto open_other_zone;
1301 * List of opened zones is per-device, shared across all threads.
1302 * Start with quasi-random candidate zone.
1303 * Ignore zones which don't belong to thread's offset/size area.
1305 open_zone_idx = pick_random_zone_idx(f, io_u);
1306 assert(!open_zone_idx ||
1307 open_zone_idx < zbdi->num_open_zones);
1308 tmp_idx = open_zone_idx;
1309 for (i = 0; i < zbdi->num_open_zones; i++) {
1312 if (tmp_idx >= zbdi->num_open_zones)
1314 tmpz = zbdi->open_zones[tmp_idx];
1315 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1316 open_zone_idx = tmp_idx;
1317 goto found_candidate_zone;
1323 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1324 __func__, f->file_name);
1325 pthread_mutex_unlock(&zbdi->mutex);
1330 found_candidate_zone:
1331 new_zone_idx = zbdi->open_zones[open_zone_idx];
1332 if (new_zone_idx == zone_idx)
1334 zone_idx = new_zone_idx;
1335 pthread_mutex_unlock(&zbdi->mutex);
1340 /* Both z->mutex and zbdi->mutex are held. */
1343 if (z->wp + min_bs <= zbd_zone_capacity_end(z)) {
1344 pthread_mutex_unlock(&zbdi->mutex);
1349 /* Check if number of open zones reaches one of limits. */
1351 zbdi->num_open_zones == f->max_zone - f->min_zone ||
1352 (zbdi->max_open_zones &&
1353 zbdi->num_open_zones == zbdi->max_open_zones) ||
1354 (td->o.job_max_open_zones &&
1355 td->num_open_zones == td->o.job_max_open_zones);
1357 pthread_mutex_unlock(&zbdi->mutex);
1359 /* Only z->mutex is held. */
1362 * When number of open zones reaches to one of limits, wait for
1363 * zone close before opening a new zone.
1365 if (wait_zone_close) {
1366 dprint(FD_ZBD, "%s(%s): quiesce to allow open zones to close\n",
1367 __func__, f->file_name);
1372 /* Zone 'z' is full, so try to open a new zone. */
1373 for (i = f->io_size / zbdi->zone_size; i > 0; i--) {
1378 if (!is_valid_offset(f, z->start)) {
1380 zone_idx = f->min_zone;
1381 z = get_zone(f, zone_idx);
1383 assert(is_valid_offset(f, z->start));
1386 zone_lock(td, f, z);
1389 if (zbd_open_zone(td, f, zone_idx))
1393 /* Only z->mutex is held. */
1395 /* Check whether the write fits in any of the already opened zones. */
1396 pthread_mutex_lock(&zbdi->mutex);
1397 for (i = 0; i < zbdi->num_open_zones; i++) {
1398 zone_idx = zbdi->open_zones[i];
1399 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1401 pthread_mutex_unlock(&zbdi->mutex);
1404 z = get_zone(f, zone_idx);
1406 zone_lock(td, f, z);
1407 if (z->wp + min_bs <= zbd_zone_capacity_end(z))
1409 pthread_mutex_lock(&zbdi->mutex);
1413 * When any I/O is in-flight or when all I/Os in-flight get completed,
1414 * the I/Os might have closed zones then retry the steps to open a zone.
1415 * Before retry, call io_u_quiesce() to complete in-flight writes.
1417 in_flight = any_io_in_flight();
1418 if (in_flight || should_retry) {
1419 dprint(FD_ZBD, "%s(%s): wait zone close and retry open zones\n",
1420 __func__, f->file_name);
1421 pthread_mutex_unlock(&zbdi->mutex);
1424 zone_lock(td, f, z);
1425 should_retry = in_flight;
1429 pthread_mutex_unlock(&zbdi->mutex);
1431 dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__,
1436 dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name,
1438 io_u->offset = z->start;
1440 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1444 /* The caller must hold z->mutex. */
1445 static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td,
1447 struct fio_zone_info *z)
1449 const struct fio_file *f = io_u->file;
1450 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
1452 if (!zbd_open_zone(td, f, zbd_zone_nr(f, z))) {
1454 z = zbd_convert_to_open_zone(td, io_u);
1458 if (z->verify_block * min_bs >= z->capacity) {
1459 log_err("%s: %d * %"PRIu64" >= %"PRIu64"\n", f->file_name, z->verify_block,
1460 min_bs, z->capacity);
1462 * If the assertion below fails during a test run, adding
1463 * "--experimental_verify=1" to the command line may help.
1467 io_u->offset = z->start + z->verify_block * min_bs;
1468 if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1469 log_err("%s: %llu + %llu >= %"PRIu64"\n", f->file_name, io_u->offset,
1470 io_u->buflen, zbd_zone_capacity_end(z));
1473 z->verify_block += io_u->buflen / min_bs;
1479 * Find another zone which has @min_bytes of readable data. Search in zones
1480 * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf.
1482 * Either returns NULL or returns a zone pointer. When the zone has write
1483 * pointer, hold the mutex for the zone.
1485 static struct fio_zone_info *
1486 zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes,
1487 struct fio_zone_info *zb, struct fio_zone_info *zl)
1489 struct fio_file *f = io_u->file;
1490 struct fio_zone_info *z1, *z2;
1491 const struct fio_zone_info *const zf = get_zone(f, f->min_zone);
1494 * Skip to the next non-empty zone in case of sequential I/O and to
1495 * the nearest non-empty zone in case of random I/O.
1497 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1498 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1500 zone_lock(td, f, z1);
1501 if (z1->start + min_bytes <= z1->wp)
1505 } else if (!td_random(td)) {
1508 if (td_random(td) && z2 >= zf &&
1509 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1511 zone_lock(td, f, z2);
1512 if (z2->start + min_bytes <= z2->wp)
1518 dprint(FD_ZBD, "%s: no zone has %"PRIu64" bytes of readable data\n",
1519 f->file_name, min_bytes);
1524 * zbd_end_zone_io - update zone status at command completion
1526 * @z: zone info pointer
1528 * If the write command made the zone full, close it.
1530 * The caller must hold z->mutex.
1532 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1533 struct fio_zone_info *z)
1535 const struct fio_file *f = io_u->file;
1537 if (io_u->ddir == DDIR_WRITE &&
1538 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1539 pthread_mutex_lock(&f->zbd_info->mutex);
1540 zbd_close_zone(td, f, zbd_zone_nr(f, z));
1541 pthread_mutex_unlock(&f->zbd_info->mutex);
1546 * zbd_queue_io - update the write pointer of a sequential zone
1548 * @success: Whether or not the I/O unit has been queued successfully
1549 * @q: queueing status (busy, completed or queued).
1551 * For write and trim operations, update the write pointer of the I/O unit
1554 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1557 const struct fio_file *f = io_u->file;
1558 struct zoned_block_device_info *zbd_info = f->zbd_info;
1559 struct fio_zone_info *z;
1565 zone_idx = zbd_zone_idx(f, io_u->offset);
1566 assert(zone_idx < zbd_info->nr_zones);
1567 z = get_zone(f, zone_idx);
1575 "%s: queued I/O (%lld, %llu) for zone %u\n",
1576 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1578 switch (io_u->ddir) {
1580 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1581 zbd_zone_capacity_end(z));
1582 pthread_mutex_lock(&zbd_info->mutex);
1584 * z->wp > zone_end means that one or more I/O errors
1587 if (z->wp <= zone_end) {
1588 zbd_info->sectors_with_data += zone_end - z->wp;
1589 zbd_info->wp_sectors_with_data += zone_end - z->wp;
1591 pthread_mutex_unlock(&zbd_info->mutex);
1598 if (q == FIO_Q_COMPLETED && !io_u->error)
1599 zbd_end_zone_io(td, io_u, z);
1602 if (!success || q != FIO_Q_QUEUED) {
1603 /* BUSY or COMPLETED: unlock the zone */
1605 io_u->zbd_put_io = NULL;
1610 * zbd_put_io - Unlock an I/O unit target zone lock
1613 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1615 const struct fio_file *f = io_u->file;
1616 struct zoned_block_device_info *zbd_info = f->zbd_info;
1617 struct fio_zone_info *z;
1622 zone_idx = zbd_zone_idx(f, io_u->offset);
1623 assert(zone_idx < zbd_info->nr_zones);
1624 z = get_zone(f, zone_idx);
1629 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1630 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1632 zbd_end_zone_io(td, io_u, z);
1635 zbd_check_swd(td, f);
1639 * Windows and MacOS do not define this.
1642 #define EREMOTEIO 121 /* POSIX value */
1645 bool zbd_unaligned_write(int error_code)
1647 switch (error_code) {
1656 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1657 * @td: FIO thread data.
1658 * @io_u: FIO I/O unit.
1660 * For sequential workloads, change the file offset to skip zoneskip bytes when
1661 * no more IO can be performed in the current zone.
1662 * - For read workloads, zoneskip is applied when the io has reached the end of
1663 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1664 * - For write workloads, zoneskip is applied when the zone is full.
1665 * This applies only to read and write operations.
1667 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1669 struct fio_file *f = io_u->file;
1670 enum fio_ddir ddir = io_u->ddir;
1671 struct fio_zone_info *z;
1674 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1675 assert(td->o.zone_size);
1676 assert(f->zbd_info);
1678 zone_idx = zbd_zone_idx(f, f->last_pos[ddir]);
1679 z = get_zone(f, zone_idx);
1682 * When the zone capacity is smaller than the zone size and the I/O is
1683 * sequential write, skip to zone end if the latest position is at the
1684 * zone capacity limit.
1686 if (z->capacity < f->zbd_info->zone_size && !td_random(td) &&
1687 ddir == DDIR_WRITE &&
1688 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1690 "%s: Jump from zone capacity limit to zone end:"
1691 " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n",
1692 f->file_name, f->last_pos[ddir],
1693 zbd_zone_end(z), zone_idx, z->capacity);
1694 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1695 f->last_pos[ddir] = zbd_zone_end(z);
1699 * zone_skip is valid only for sequential workloads.
1701 if (td_random(td) || !td->o.zone_skip)
1705 * It is time to switch to a new zone if:
1706 * - zone_bytes == zone_size bytes have already been accessed
1707 * - The last position reached the end of the current zone.
1708 * - For reads with td->o.read_beyond_wp == false, the last position
1709 * reached the zone write pointer.
1711 if (td->zone_bytes >= td->o.zone_size ||
1712 f->last_pos[ddir] >= zbd_zone_end(z) ||
1713 (ddir == DDIR_READ &&
1714 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1719 f->file_offset += td->o.zone_size + td->o.zone_skip;
1722 * Wrap from the beginning, if we exceed the file size
1724 if (f->file_offset >= f->real_file_size)
1725 f->file_offset = get_start_offset(td, f);
1727 f->last_pos[ddir] = f->file_offset;
1728 td->io_skip_bytes += td->o.zone_skip;
1733 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1735 * @td: FIO thread data.
1736 * @io_u: FIO I/O unit.
1737 * @ddir: I/O direction before adjustment.
1739 * Return adjusted I/O direction.
1741 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1745 * In case read direction is chosen for the first random I/O, fio with
1746 * zonemode=zbd stops because no data can be read from zoned block
1747 * devices with all empty zones. Overwrite the first I/O direction as
1748 * write to make sure data to read exists.
1750 assert(io_u->file->zbd_info);
1751 if (ddir != DDIR_READ || !td_rw(td))
1754 if (io_u->file->zbd_info->sectors_with_data ||
1755 td->o.read_beyond_wp)
1762 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1763 * @td: FIO thread data.
1764 * @io_u: FIO I/O unit.
1766 * Locking strategy: returns with z->mutex locked if and only if z refers
1767 * to a sequential zone and if io_u_accept is returned. z is the zone that
1768 * corresponds to io_u->offset at the end of this function.
1770 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1772 struct fio_file *f = io_u->file;
1773 struct zoned_block_device_info *zbdi = f->zbd_info;
1774 uint32_t zone_idx_b;
1775 struct fio_zone_info *zb, *zl, *orig_zb;
1776 uint32_t orig_len = io_u->buflen;
1777 uint64_t min_bs = td->o.min_bs[io_u->ddir];
1783 assert(is_valid_offset(f, io_u->offset));
1784 assert(io_u->buflen);
1785 zone_idx_b = zbd_zone_idx(f, io_u->offset);
1786 zb = get_zone(f, zone_idx_b);
1790 /* Accept non-write I/Os for conventional zones. */
1791 if (io_u->ddir != DDIR_WRITE)
1794 * Make sure that writes to conventional zones
1795 * don't cross over to any sequential zones.
1797 if (!(zb + 1)->has_wp ||
1798 io_u->offset + io_u->buflen <= (zb + 1)->start)
1801 if (io_u->offset + min_bs > (zb + 1)->start) {
1803 "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n",
1804 f->file_name, io_u->offset,
1805 min_bs, (zb + 1)->start);
1806 io_u->offset = zb->start + (zb + 1)->start - io_u->offset;
1807 new_len = min(io_u->buflen, (zb + 1)->start - io_u->offset);
1809 new_len = (zb + 1)->start - io_u->offset;
1811 io_u->buflen = new_len / min_bs * min_bs;
1816 * Accept the I/O offset for reads if reading beyond the write pointer
1819 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1820 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1823 zbd_check_swd(td, f);
1825 zone_lock(td, f, zb);
1827 switch (io_u->ddir) {
1829 if (td->runstate == TD_VERIFYING && td_write(td)) {
1830 zb = zbd_replay_write_order(td, io_u, zb);
1834 * Check that there is enough written data in the zone to do an
1835 * I/O of at least min_bs B. If there isn't, find a new zone for
1838 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1839 zb->wp - zb->start : 0;
1840 if (range < min_bs ||
1841 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1843 zl = get_zone(f, f->max_zone);
1844 zb = zbd_find_zone(td, io_u, min_bs, zb, zl);
1847 "%s: zbd_find_zone(%lld, %llu) failed\n",
1848 f->file_name, io_u->offset,
1853 * zbd_find_zone() returned a zone with a range of at
1856 range = zb->wp - zb->start;
1857 assert(range >= min_bs);
1860 io_u->offset = zb->start;
1863 * Make sure the I/O is within the zone valid data range while
1864 * maximizing the I/O size and preserving randomness.
1866 if (range <= io_u->buflen)
1867 io_u->offset = zb->start;
1868 else if (td_random(td))
1869 io_u->offset = zb->start +
1870 ((io_u->offset - orig_zb->start) %
1871 (range - io_u->buflen)) / min_bs * min_bs;
1873 * When zbd_find_zone() returns a conventional zone,
1874 * we can simply accept the new i/o offset here.
1879 * Make sure the I/O does not cross over the zone wp position.
1881 new_len = min((unsigned long long)io_u->buflen,
1882 (unsigned long long)(zb->wp - io_u->offset));
1883 new_len = new_len / min_bs * min_bs;
1884 if (new_len < io_u->buflen) {
1885 io_u->buflen = new_len;
1886 dprint(FD_IO, "Changed length from %u into %llu\n",
1887 orig_len, io_u->buflen);
1889 assert(zb->start <= io_u->offset);
1890 assert(io_u->offset + io_u->buflen <= zb->wp);
1893 if (io_u->buflen > zbdi->zone_size) {
1894 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
1896 "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n",
1897 f->file_name, io_u->buflen, zbdi->zone_size);
1900 if (!zbd_open_zone(td, f, zone_idx_b)) {
1902 zb = zbd_convert_to_open_zone(td, io_u);
1904 dprint(FD_IO, "%s: can't convert to open zone",
1909 /* Check whether the zone reset threshold has been exceeded */
1910 if (td->o.zrf.u.f) {
1911 if (zbdi->wp_sectors_with_data >=
1912 f->io_size * td->o.zrt.u.f &&
1913 zbd_dec_and_reset_write_cnt(td, f)) {
1917 /* Reset the zone pointer if necessary */
1918 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
1919 assert(td->o.verify == VERIFY_NONE);
1921 * Since previous write requests may have been submitted
1922 * asynchronously and since we will submit the zone
1923 * reset synchronously, wait until previously submitted
1924 * write requests have completed before issuing a
1929 if (zbd_reset_zone(td, f, zb) < 0)
1932 if (zb->capacity < min_bs) {
1933 td_verror(td, EINVAL, "ZCAP is less min_bs");
1934 log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n",
1935 zb->capacity, min_bs);
1939 /* Make writes occur at the write pointer */
1940 assert(!zbd_zone_full(f, zb, min_bs));
1941 io_u->offset = zb->wp;
1942 if (!is_valid_offset(f, io_u->offset)) {
1943 td_verror(td, EINVAL, "invalid WP value");
1944 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
1945 f->file_name, io_u->offset);
1949 * Make sure that the buflen is a multiple of the minimal
1950 * block size. Give up if shrinking would make the request too
1953 new_len = min((unsigned long long)io_u->buflen,
1954 zbd_zone_capacity_end(zb) - io_u->offset);
1955 new_len = new_len / min_bs * min_bs;
1956 if (new_len == io_u->buflen)
1958 if (new_len >= min_bs) {
1959 io_u->buflen = new_len;
1960 dprint(FD_IO, "Changed length from %u into %llu\n",
1961 orig_len, io_u->buflen);
1964 td_verror(td, EIO, "zone remainder too small");
1965 log_err("zone remainder %lld smaller than min block size %"PRIu64"\n",
1966 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
1969 /* Check random trim targets a non-empty zone */
1970 if (!td_random(td) || zb->wp > zb->start)
1973 /* Find out a non-empty zone to trim */
1975 zl = get_zone(f, f->max_zone);
1976 zb = zbd_find_zone(td, io_u, 1, zb, zl);
1978 io_u->offset = zb->start;
1979 dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n",
1980 f->file_name, io_u->offset);
1987 case DDIR_SYNC_FILE_RANGE:
1998 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
1999 assert(!io_u->zbd_queue_io);
2000 assert(!io_u->zbd_put_io);
2001 io_u->zbd_queue_io = zbd_queue_io;
2002 io_u->zbd_put_io = zbd_put_io;
2004 * Since we return with the zone lock still held,
2005 * add an annotation to let Coverity know that it
2008 /* coverity[missing_unlock] */
2012 if (zb && zb->has_wp)
2017 /* Return a string with ZBD statistics */
2018 char *zbd_write_status(const struct thread_stat *ts)
2022 if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0)
2028 * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim
2030 * @td: FIO thread data.
2031 * @io_u: FIO I/O unit.
2033 * It is assumed that z->mutex is already locked.
2034 * Return io_u_completed when reset zone succeeds. Return 0 when the target zone
2035 * does not have write pointer. On error, return negative errno.
2037 int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u)
2039 struct fio_file *f = io_u->file;
2040 struct fio_zone_info *z;
2044 zone_idx = zbd_zone_idx(f, io_u->offset);
2045 z = get_zone(f, zone_idx);
2050 if (io_u->offset != z->start) {
2051 log_err("Trim offset not at zone start (%lld)\n", io_u->offset);
2055 ret = zbd_reset_zone((struct thread_data *)td, f, z);
2059 return io_u_completed;