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 * zbd_open_zone - Add a zone to the array of open zones.
382 * @td: fio thread data.
383 * @f: fio file that has the open zones to add.
384 * @zone_idx: Index of the zone to add.
386 * Open a ZBD zone if it is not already open. Returns true if either the zone
387 * was already open or if the zone was successfully added to the array of open
388 * zones without exceeding the maximum number of open zones. Returns false if
389 * the zone was not already open and opening the zone would cause the zone limit
392 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
395 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
396 struct zoned_block_device_info *zbdi = f->zbd_info;
397 struct fio_zone_info *z = get_zone(f, zone_idx);
400 if (z->cond == ZBD_ZONE_COND_OFFLINE)
404 * Skip full zones with data verification enabled because resetting a
405 * zone causes data loss and hence causes verification to fail.
407 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
411 * zbdi->max_open_zones == 0 means that there is no limit on the maximum
412 * number of open zones. In this case, do no track open zones in
413 * zbdi->open_zones array.
415 if (!zbdi->max_open_zones)
418 pthread_mutex_lock(&zbdi->mutex);
422 * If the zone is going to be completely filled by writes
423 * already in-flight, handle it as a full zone instead of an
426 if (z->wp >= zbd_zone_capacity_end(z))
431 /* Zero means no limit */
432 if (td->o.job_max_open_zones > 0 &&
433 td->num_open_zones >= td->o.job_max_open_zones)
435 if (zbdi->num_open_zones >= zbdi->max_open_zones)
437 dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx);
438 zbdi->open_zones[zbdi->num_open_zones++] = zone_idx;
439 td->num_open_zones++;
444 pthread_mutex_unlock(&zbdi->mutex);
448 /* Verify whether direct I/O is used for all host-managed zoned drives. */
449 static bool zbd_using_direct_io(void)
451 struct thread_data *td;
456 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
458 for_each_file(td, f, j) {
460 f->zbd_info->model == ZBD_HOST_MANAGED)
468 /* Whether or not the I/O range for f includes one or more sequential zones */
469 static bool zbd_is_seq_job(struct fio_file *f)
471 uint32_t zone_idx, zone_idx_b, zone_idx_e;
476 zone_idx_b = zbd_zone_idx(f, f->file_offset);
477 zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1);
478 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
479 if (get_zone(f, zone_idx)->has_wp)
486 * Verify whether offset and size parameters are aligned with zone boundaries.
488 static bool zbd_verify_sizes(void)
490 const struct fio_zone_info *z;
491 struct thread_data *td;
493 uint64_t new_offset, new_end;
498 for_each_file(td, f, j) {
501 if (f->file_offset >= f->real_file_size)
503 if (!zbd_is_seq_job(f))
506 if (!td->o.zone_size) {
507 td->o.zone_size = f->zbd_info->zone_size;
508 if (!td->o.zone_size) {
509 log_err("%s: invalid 0 zone size\n",
513 } else if (td->o.zone_size != f->zbd_info->zone_size) {
514 log_err("%s: job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
515 f->file_name, td->o.zone_size,
516 f->zbd_info->zone_size);
520 if (td->o.zone_skip % td->o.zone_size) {
521 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
522 f->file_name, td->o.zone_skip,
527 zone_idx = zbd_zone_idx(f, f->file_offset);
528 z = get_zone(f, zone_idx);
529 if ((f->file_offset != z->start) &&
530 (td->o.td_ddir != TD_DDIR_READ)) {
531 new_offset = zbd_zone_end(z);
532 if (new_offset >= f->file_offset + f->io_size) {
533 log_info("%s: io_size must be at least one zone\n",
537 log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n",
538 f->file_name, f->file_offset,
540 f->io_size -= (new_offset - f->file_offset);
541 f->file_offset = new_offset;
543 zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size);
544 z = get_zone(f, zone_idx);
546 if ((td->o.td_ddir != TD_DDIR_READ) &&
547 (f->file_offset + f->io_size != new_end)) {
548 if (new_end <= f->file_offset) {
549 log_info("%s: io_size must be at least one zone\n",
553 log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n",
554 f->file_name, f->io_size,
555 new_end - f->file_offset);
556 f->io_size = new_end - f->file_offset;
564 static bool zbd_verify_bs(void)
566 struct thread_data *td;
572 (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] ||
573 td->o.bssplit_nr[DDIR_TRIM])) {
574 log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n");
577 for_each_file(td, f, j) {
582 zone_size = f->zbd_info->zone_size;
583 if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) {
584 log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n",
585 f->file_name, td->o.bs[DDIR_TRIM],
589 for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) {
590 if (td->o.verify != VERIFY_NONE &&
591 zone_size % td->o.bs[k] != 0) {
592 log_info("%s: block size %llu is not a divisor of the zone size %"PRIu64"\n",
593 f->file_name, td->o.bs[k],
603 static int ilog2(uint64_t i)
615 * Initialize f->zbd_info for devices that are not zoned block devices. This
616 * allows to execute a ZBD workload against a non-ZBD device.
618 static int init_zone_info(struct thread_data *td, struct fio_file *f)
621 struct fio_zone_info *p;
622 uint64_t zone_size = td->o.zone_size;
623 uint64_t zone_capacity = td->o.zone_capacity;
624 struct zoned_block_device_info *zbd_info = NULL;
627 if (zone_size == 0) {
628 log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n",
633 if (zone_size < 512) {
634 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
639 if (zone_capacity == 0)
640 zone_capacity = zone_size;
642 if (zone_capacity > zone_size) {
643 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
644 f->file_name, td->o.zone_capacity, td->o.zone_size);
648 if (f->real_file_size < zone_size) {
649 log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n",
650 f->file_name, f->real_file_size, zone_size);
654 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
655 zbd_info = scalloc(1, sizeof(*zbd_info) +
656 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
660 mutex_init_pshared(&zbd_info->mutex);
661 zbd_info->refcount = 1;
662 p = &zbd_info->zone_info[0];
663 for (i = 0; i < nr_zones; i++, p++) {
664 mutex_init_pshared_with_type(&p->mutex,
665 PTHREAD_MUTEX_RECURSIVE);
666 p->start = i * zone_size;
668 p->type = ZBD_ZONE_TYPE_SWR;
669 p->cond = ZBD_ZONE_COND_EMPTY;
670 p->capacity = zone_capacity;
674 p->start = nr_zones * zone_size;
676 f->zbd_info = zbd_info;
677 f->zbd_info->zone_size = zone_size;
678 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
679 ilog2(zone_size) : 0;
680 f->zbd_info->nr_zones = nr_zones;
685 * Maximum number of zones to report in one operation.
687 #define ZBD_REPORT_MAX_ZONES 8192U
690 * Parse the device zone report and store it in f->zbd_info. Must be called
691 * only for devices that are zoned, namely those with a model != ZBD_NONE.
693 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
696 struct zbd_zone *zones, *z;
697 struct fio_zone_info *p;
698 uint64_t zone_size, offset;
699 struct zoned_block_device_info *zbd_info = NULL;
700 int i, j, ret = -ENOMEM;
702 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
706 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
709 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
714 zone_size = zones[0].len;
715 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
717 if (td->o.zone_size == 0) {
718 td->o.zone_size = zone_size;
719 } else if (td->o.zone_size != zone_size) {
720 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
721 f->file_name, td->o.zone_size, zone_size);
726 dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n", f->file_name,
727 nr_zones, zone_size / 1024);
729 zbd_info = scalloc(1, sizeof(*zbd_info) +
730 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
733 mutex_init_pshared(&zbd_info->mutex);
734 zbd_info->refcount = 1;
735 p = &zbd_info->zone_info[0];
736 for (offset = 0, j = 0; j < nr_zones;) {
738 for (i = 0; i < nrz; i++, j++, z++, p++) {
739 mutex_init_pshared_with_type(&p->mutex,
740 PTHREAD_MUTEX_RECURSIVE);
742 p->capacity = z->capacity;
744 case ZBD_ZONE_COND_NOT_WP:
745 case ZBD_ZONE_COND_FULL:
746 p->wp = p->start + p->capacity;
749 assert(z->start <= z->wp);
750 assert(z->wp <= z->start + zone_size);
756 case ZBD_ZONE_TYPE_SWR:
765 if (j > 0 && p->start != p[-1].start + zone_size) {
766 log_info("%s: invalid zone data\n",
773 offset = z->start + z->len;
776 nrz = zbd_report_zones(td, f, offset, zones,
777 min((uint32_t)(nr_zones - j),
778 ZBD_REPORT_MAX_ZONES));
781 log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n",
782 offset, f->file_name, -ret);
788 zbd_info->zone_info[nr_zones].start = offset;
790 f->zbd_info = zbd_info;
791 f->zbd_info->zone_size = zone_size;
792 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
793 ilog2(zone_size) : 0;
794 f->zbd_info->nr_zones = nr_zones;
804 static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f)
806 struct zoned_block_device_info *zbd = f->zbd_info;
807 unsigned int max_open_zones;
810 if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) {
811 /* Only host-managed devices have a max open limit */
812 zbd->max_open_zones = td->o.max_open_zones;
816 /* If host-managed, get the max open limit */
817 ret = zbd_get_max_open_zones(td, f, &max_open_zones);
821 if (!max_open_zones) {
822 /* No device limit */
823 zbd->max_open_zones = td->o.max_open_zones;
824 } else if (!td->o.max_open_zones) {
825 /* No user limit. Set limit to device limit */
826 zbd->max_open_zones = max_open_zones;
827 } else if (td->o.max_open_zones <= max_open_zones) {
828 /* Both user limit and dev limit. User limit not too large */
829 zbd->max_open_zones = td->o.max_open_zones;
831 /* Both user limit and dev limit. User limit too large */
832 td_verror(td, EINVAL,
833 "Specified --max_open_zones is too large");
834 log_err("Specified --max_open_zones (%d) is larger than max (%u)\n",
835 td->o.max_open_zones, max_open_zones);
840 /* Ensure that the limit is not larger than FIO's internal limit */
841 if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) {
842 td_verror(td, EINVAL, "'max_open_zones' value is too large");
843 log_err("'max_open_zones' value is larger than %u\n", ZBD_MAX_OPEN_ZONES);
847 dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n",
848 f->file_name, zbd->max_open_zones);
854 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
856 * Returns 0 upon success and a negative error code upon failure.
858 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
860 enum zbd_zoned_model zbd_model;
863 assert(td->o.zone_mode == ZONE_MODE_ZBD);
865 ret = zbd_get_zoned_model(td, f, &zbd_model);
871 case ZBD_HOST_MANAGED:
872 ret = parse_zone_info(td, f);
877 ret = init_zone_info(td, f);
882 td_verror(td, EINVAL, "Unsupported zoned model");
883 log_err("Unsupported zoned model\n");
888 f->zbd_info->model = zbd_model;
890 ret = zbd_set_max_open_zones(td, f);
892 zbd_free_zone_info(f);
899 void zbd_free_zone_info(struct fio_file *f)
905 pthread_mutex_lock(&f->zbd_info->mutex);
906 refcount = --f->zbd_info->refcount;
907 pthread_mutex_unlock(&f->zbd_info->mutex);
909 assert((int32_t)refcount >= 0);
916 * Initialize f->zbd_info.
918 * Returns 0 upon success and a negative error code upon failure.
920 * Note: this function can only work correctly if it is called before the first
923 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
925 struct thread_data *td2;
929 for_each_td(td2, i) {
930 for_each_file(td2, f2, j) {
931 if (td2 == td && f2 == file)
934 strcmp(f2->file_name, file->file_name) != 0)
936 file->zbd_info = f2->zbd_info;
937 file->zbd_info->refcount++;
942 ret = zbd_create_zone_info(td, file);
944 td_verror(td, -ret, "zbd_create_zone_info() failed");
948 int zbd_init_files(struct thread_data *td)
953 for_each_file(td, f, i) {
954 if (zbd_init_zone_info(td, f))
960 void zbd_recalc_options_with_zone_granularity(struct thread_data *td)
965 for_each_file(td, f, i) {
966 struct zoned_block_device_info *zbd = f->zbd_info;
967 // zonemode=strided doesn't get per-file zone size.
968 uint64_t zone_size = zbd ? zbd->zone_size : td->o.zone_size;
973 if (td->o.size_nz > 0) {
974 td->o.size = td->o.size_nz * zone_size;
976 if (td->o.io_size_nz > 0) {
977 td->o.io_size = td->o.io_size_nz * zone_size;
979 if (td->o.start_offset_nz > 0) {
980 td->o.start_offset = td->o.start_offset_nz * zone_size;
982 if (td->o.offset_increment_nz > 0) {
983 td->o.offset_increment = td->o.offset_increment_nz * zone_size;
985 if (td->o.zone_skip_nz > 0) {
986 td->o.zone_skip = td->o.zone_skip_nz * zone_size;
991 int zbd_setup_files(struct thread_data *td)
996 if (!zbd_using_direct_io()) {
997 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
1001 if (!zbd_verify_sizes())
1004 if (!zbd_verify_bs())
1007 for_each_file(td, f, i) {
1008 struct zoned_block_device_info *zbd = f->zbd_info;
1009 struct fio_zone_info *z;
1014 f->min_zone = zbd_zone_idx(f, f->file_offset);
1015 f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size);
1018 * When all zones in the I/O range are conventional, io_size
1019 * can be smaller than zone size, making min_zone the same
1020 * as max_zone. This is why the assert below needs to be made
1023 if (zbd_is_seq_job(f))
1024 assert(f->min_zone < f->max_zone);
1026 if (td->o.max_open_zones > 0 &&
1027 zbd->max_open_zones != td->o.max_open_zones) {
1028 log_err("Different 'max_open_zones' values\n");
1033 * The per job max open zones limit cannot be used without a
1034 * global max open zones limit. (As the tracking of open zones
1035 * is disabled when there is no global max open zones limit.)
1037 if (td->o.job_max_open_zones && !zbd->max_open_zones) {
1038 log_err("'job_max_open_zones' cannot be used without a global open zones limit\n");
1043 * zbd->max_open_zones is the global limit shared for all jobs
1044 * that target the same zoned block device. Force sync the per
1045 * thread global limit with the actual global limit. (The real
1046 * per thread/job limit is stored in td->o.job_max_open_zones).
1048 td->o.max_open_zones = zbd->max_open_zones;
1050 for (zi = f->min_zone; zi < f->max_zone; zi++) {
1051 z = &zbd->zone_info[zi];
1052 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
1053 z->cond != ZBD_ZONE_COND_EXP_OPEN)
1055 if (zbd_open_zone(td, f, zi))
1058 * If the number of open zones exceeds specified limits,
1059 * reset all extra open zones.
1061 if (zbd_reset_zone(td, f, z) < 0) {
1062 log_err("Failed to reest zone %d\n", zi);
1072 * Reset zbd_info.write_cnt, the counter that counts down towards the next
1075 static void _zbd_reset_write_cnt(const struct thread_data *td,
1076 const struct fio_file *f)
1078 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
1080 f->zbd_info->write_cnt = td->o.zrf.u.f ?
1081 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
1084 static void zbd_reset_write_cnt(const struct thread_data *td,
1085 const struct fio_file *f)
1087 pthread_mutex_lock(&f->zbd_info->mutex);
1088 _zbd_reset_write_cnt(td, f);
1089 pthread_mutex_unlock(&f->zbd_info->mutex);
1092 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
1093 const struct fio_file *f)
1095 uint32_t write_cnt = 0;
1097 pthread_mutex_lock(&f->zbd_info->mutex);
1098 assert(f->zbd_info->write_cnt);
1099 if (f->zbd_info->write_cnt)
1100 write_cnt = --f->zbd_info->write_cnt;
1102 _zbd_reset_write_cnt(td, f);
1103 pthread_mutex_unlock(&f->zbd_info->mutex);
1105 return write_cnt == 0;
1113 /* Calculate the number of sectors with data (swd) and perform action 'a' */
1114 static uint64_t zbd_process_swd(struct thread_data *td,
1115 const struct fio_file *f, enum swd_action a)
1117 struct fio_zone_info *zb, *ze, *z;
1119 uint64_t wp_swd = 0;
1121 zb = get_zone(f, f->min_zone);
1122 ze = get_zone(f, f->max_zone);
1123 for (z = zb; z < ze; z++) {
1125 zone_lock(td, f, z);
1126 wp_swd += z->wp - z->start;
1128 swd += z->wp - z->start;
1130 pthread_mutex_lock(&f->zbd_info->mutex);
1133 assert(f->zbd_info->sectors_with_data == swd);
1134 assert(f->zbd_info->wp_sectors_with_data == wp_swd);
1137 f->zbd_info->sectors_with_data = swd;
1138 f->zbd_info->wp_sectors_with_data = wp_swd;
1141 pthread_mutex_unlock(&f->zbd_info->mutex);
1142 for (z = zb; z < ze; z++)
1150 * The swd check is useful for debugging but takes too much time to leave
1151 * it enabled all the time. Hence it is disabled by default.
1153 static const bool enable_check_swd = false;
1155 /* Check whether the values of zbd_info.*sectors_with_data are correct. */
1156 static void zbd_check_swd(struct thread_data *td, const struct fio_file *f)
1158 if (!enable_check_swd)
1161 zbd_process_swd(td, f, CHECK_SWD);
1164 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
1166 struct fio_zone_info *zb, *ze;
1169 if (!f->zbd_info || !td_write(td))
1172 zb = get_zone(f, f->min_zone);
1173 ze = get_zone(f, f->max_zone);
1174 swd = zbd_process_swd(td, f, SET_SWD);
1175 dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name,
1178 * If data verification is enabled reset the affected zones before
1179 * writing any data to avoid that a zone reset has to be issued while
1180 * writing data, which causes data loss.
1182 if (td->o.verify != VERIFY_NONE && td->runstate != TD_VERIFYING)
1183 zbd_reset_zones(td, f, zb, ze);
1184 zbd_reset_write_cnt(td, f);
1187 /* Return random zone index for one of the open zones. */
1188 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1189 const struct io_u *io_u)
1191 return (io_u->offset - f->file_offset) * f->zbd_info->num_open_zones /
1195 static bool any_io_in_flight(void)
1197 struct thread_data *td;
1200 for_each_td(td, i) {
1201 if (td->io_u_in_flight)
1209 * Modify the offset of an I/O unit that does not refer to an open zone such
1210 * that it refers to an open zone. Close an open zone and open a new zone if
1211 * necessary. The open zone is searched across sequential zones.
1212 * This algorithm can only work correctly if all write pointers are
1213 * a multiple of the fio block size. The caller must neither hold z->mutex
1214 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1216 static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
1219 const uint64_t min_bs = td->o.min_bs[io_u->ddir];
1220 struct fio_file *f = io_u->file;
1221 struct zoned_block_device_info *zbdi = f->zbd_info;
1222 struct fio_zone_info *z;
1223 unsigned int open_zone_idx = -1;
1224 uint32_t zone_idx, new_zone_idx;
1226 bool wait_zone_close;
1228 bool should_retry = true;
1230 assert(is_valid_offset(f, io_u->offset));
1232 if (zbdi->max_open_zones || td->o.job_max_open_zones) {
1234 * This statement accesses zbdi->open_zones[] on purpose
1237 zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)];
1239 zone_idx = zbd_zone_idx(f, io_u->offset);
1241 if (zone_idx < f->min_zone)
1242 zone_idx = f->min_zone;
1243 else if (zone_idx >= f->max_zone)
1244 zone_idx = f->max_zone - 1;
1245 dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1246 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1249 * Since z->mutex is the outer lock and zbdi->mutex the inner
1250 * lock it can happen that the state of the zone with index zone_idx
1251 * has changed after 'z' has been assigned and before zbdi->mutex
1252 * has been obtained. Hence the loop.
1257 z = get_zone(f, zone_idx);
1259 zone_lock(td, f, z);
1260 pthread_mutex_lock(&zbdi->mutex);
1262 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1263 zbdi->max_open_zones == 0 && td->o.job_max_open_zones == 0)
1265 if (zbdi->num_open_zones == 0) {
1266 dprint(FD_ZBD, "%s(%s): no zones are open\n",
1267 __func__, f->file_name);
1268 goto open_other_zone;
1273 * List of opened zones is per-device, shared across all threads.
1274 * Start with quasi-random candidate zone.
1275 * Ignore zones which don't belong to thread's offset/size area.
1277 open_zone_idx = pick_random_zone_idx(f, io_u);
1278 assert(!open_zone_idx ||
1279 open_zone_idx < zbdi->num_open_zones);
1280 tmp_idx = open_zone_idx;
1281 for (i = 0; i < zbdi->num_open_zones; i++) {
1284 if (tmp_idx >= zbdi->num_open_zones)
1286 tmpz = zbdi->open_zones[tmp_idx];
1287 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1288 open_zone_idx = tmp_idx;
1289 goto found_candidate_zone;
1295 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1296 __func__, f->file_name);
1297 pthread_mutex_unlock(&zbdi->mutex);
1302 found_candidate_zone:
1303 new_zone_idx = zbdi->open_zones[open_zone_idx];
1304 if (new_zone_idx == zone_idx)
1306 zone_idx = new_zone_idx;
1307 pthread_mutex_unlock(&zbdi->mutex);
1312 /* Both z->mutex and zbdi->mutex are held. */
1315 if (z->wp + min_bs <= zbd_zone_capacity_end(z)) {
1316 pthread_mutex_unlock(&zbdi->mutex);
1321 /* Check if number of open zones reaches one of limits. */
1323 zbdi->num_open_zones == f->max_zone - f->min_zone ||
1324 (zbdi->max_open_zones &&
1325 zbdi->num_open_zones == zbdi->max_open_zones) ||
1326 (td->o.job_max_open_zones &&
1327 td->num_open_zones == td->o.job_max_open_zones);
1329 pthread_mutex_unlock(&zbdi->mutex);
1331 /* Only z->mutex is held. */
1334 * When number of open zones reaches to one of limits, wait for
1335 * zone close before opening a new zone.
1337 if (wait_zone_close) {
1338 dprint(FD_ZBD, "%s(%s): quiesce to allow open zones to close\n",
1339 __func__, f->file_name);
1344 /* Zone 'z' is full, so try to open a new zone. */
1345 for (i = f->io_size / zbdi->zone_size; i > 0; i--) {
1350 if (!is_valid_offset(f, z->start)) {
1352 zone_idx = f->min_zone;
1353 z = get_zone(f, zone_idx);
1355 assert(is_valid_offset(f, z->start));
1358 zone_lock(td, f, z);
1361 if (zbd_open_zone(td, f, zone_idx))
1365 /* Only z->mutex is held. */
1367 /* Check whether the write fits in any of the already opened zones. */
1368 pthread_mutex_lock(&zbdi->mutex);
1369 for (i = 0; i < zbdi->num_open_zones; i++) {
1370 zone_idx = zbdi->open_zones[i];
1371 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1373 pthread_mutex_unlock(&zbdi->mutex);
1376 z = get_zone(f, zone_idx);
1378 zone_lock(td, f, z);
1379 if (z->wp + min_bs <= zbd_zone_capacity_end(z))
1381 pthread_mutex_lock(&zbdi->mutex);
1385 * When any I/O is in-flight or when all I/Os in-flight get completed,
1386 * the I/Os might have closed zones then retry the steps to open a zone.
1387 * Before retry, call io_u_quiesce() to complete in-flight writes.
1389 in_flight = any_io_in_flight();
1390 if (in_flight || should_retry) {
1391 dprint(FD_ZBD, "%s(%s): wait zone close and retry open zones\n",
1392 __func__, f->file_name);
1393 pthread_mutex_unlock(&zbdi->mutex);
1396 zone_lock(td, f, z);
1397 should_retry = in_flight;
1401 pthread_mutex_unlock(&zbdi->mutex);
1403 dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__,
1408 dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name,
1410 io_u->offset = z->start;
1412 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1416 /* The caller must hold z->mutex. */
1417 static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td,
1419 struct fio_zone_info *z)
1421 const struct fio_file *f = io_u->file;
1422 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
1424 if (!zbd_open_zone(td, f, zbd_zone_nr(f, z))) {
1426 z = zbd_convert_to_open_zone(td, io_u);
1430 if (z->verify_block * min_bs >= z->capacity) {
1431 log_err("%s: %d * %"PRIu64" >= %"PRIu64"\n", f->file_name, z->verify_block,
1432 min_bs, z->capacity);
1434 * If the assertion below fails during a test run, adding
1435 * "--experimental_verify=1" to the command line may help.
1439 io_u->offset = z->start + z->verify_block * min_bs;
1440 if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1441 log_err("%s: %llu + %llu >= %"PRIu64"\n", f->file_name, io_u->offset,
1442 io_u->buflen, zbd_zone_capacity_end(z));
1445 z->verify_block += io_u->buflen / min_bs;
1451 * Find another zone which has @min_bytes of readable data. Search in zones
1452 * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf.
1454 * Either returns NULL or returns a zone pointer. When the zone has write
1455 * pointer, hold the mutex for the zone.
1457 static struct fio_zone_info *
1458 zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes,
1459 struct fio_zone_info *zb, struct fio_zone_info *zl)
1461 struct fio_file *f = io_u->file;
1462 struct fio_zone_info *z1, *z2;
1463 const struct fio_zone_info *const zf = get_zone(f, f->min_zone);
1466 * Skip to the next non-empty zone in case of sequential I/O and to
1467 * the nearest non-empty zone in case of random I/O.
1469 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1470 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1472 zone_lock(td, f, z1);
1473 if (z1->start + min_bytes <= z1->wp)
1477 } else if (!td_random(td)) {
1480 if (td_random(td) && z2 >= zf &&
1481 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1483 zone_lock(td, f, z2);
1484 if (z2->start + min_bytes <= z2->wp)
1490 dprint(FD_ZBD, "%s: no zone has %"PRIu64" bytes of readable data\n",
1491 f->file_name, min_bytes);
1496 * zbd_end_zone_io - update zone status at command completion
1498 * @z: zone info pointer
1500 * If the write command made the zone full, close it.
1502 * The caller must hold z->mutex.
1504 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1505 struct fio_zone_info *z)
1507 const struct fio_file *f = io_u->file;
1509 if (io_u->ddir == DDIR_WRITE &&
1510 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1511 pthread_mutex_lock(&f->zbd_info->mutex);
1512 zbd_close_zone(td, f, zbd_zone_nr(f, z));
1513 pthread_mutex_unlock(&f->zbd_info->mutex);
1518 * zbd_queue_io - update the write pointer of a sequential zone
1520 * @success: Whether or not the I/O unit has been queued successfully
1521 * @q: queueing status (busy, completed or queued).
1523 * For write and trim operations, update the write pointer of the I/O unit
1526 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1529 const struct fio_file *f = io_u->file;
1530 struct zoned_block_device_info *zbd_info = f->zbd_info;
1531 struct fio_zone_info *z;
1537 zone_idx = zbd_zone_idx(f, io_u->offset);
1538 assert(zone_idx < zbd_info->nr_zones);
1539 z = get_zone(f, zone_idx);
1547 "%s: queued I/O (%lld, %llu) for zone %u\n",
1548 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1550 switch (io_u->ddir) {
1552 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1553 zbd_zone_capacity_end(z));
1554 pthread_mutex_lock(&zbd_info->mutex);
1556 * z->wp > zone_end means that one or more I/O errors
1559 if (z->wp <= zone_end) {
1560 zbd_info->sectors_with_data += zone_end - z->wp;
1561 zbd_info->wp_sectors_with_data += zone_end - z->wp;
1563 pthread_mutex_unlock(&zbd_info->mutex);
1570 if (q == FIO_Q_COMPLETED && !io_u->error)
1571 zbd_end_zone_io(td, io_u, z);
1574 if (!success || q != FIO_Q_QUEUED) {
1575 /* BUSY or COMPLETED: unlock the zone */
1577 io_u->zbd_put_io = NULL;
1582 * zbd_put_io - Unlock an I/O unit target zone lock
1585 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1587 const struct fio_file *f = io_u->file;
1588 struct zoned_block_device_info *zbd_info = f->zbd_info;
1589 struct fio_zone_info *z;
1594 zone_idx = zbd_zone_idx(f, io_u->offset);
1595 assert(zone_idx < zbd_info->nr_zones);
1596 z = get_zone(f, zone_idx);
1601 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1602 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1604 zbd_end_zone_io(td, io_u, z);
1607 zbd_check_swd(td, f);
1611 * Windows and MacOS do not define this.
1614 #define EREMOTEIO 121 /* POSIX value */
1617 bool zbd_unaligned_write(int error_code)
1619 switch (error_code) {
1628 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1629 * @td: FIO thread data.
1630 * @io_u: FIO I/O unit.
1632 * For sequential workloads, change the file offset to skip zoneskip bytes when
1633 * no more IO can be performed in the current zone.
1634 * - For read workloads, zoneskip is applied when the io has reached the end of
1635 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1636 * - For write workloads, zoneskip is applied when the zone is full.
1637 * This applies only to read and write operations.
1639 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1641 struct fio_file *f = io_u->file;
1642 enum fio_ddir ddir = io_u->ddir;
1643 struct fio_zone_info *z;
1646 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1647 assert(td->o.zone_size);
1648 assert(f->zbd_info);
1650 zone_idx = zbd_zone_idx(f, f->last_pos[ddir]);
1651 z = get_zone(f, zone_idx);
1654 * When the zone capacity is smaller than the zone size and the I/O is
1655 * sequential write, skip to zone end if the latest position is at the
1656 * zone capacity limit.
1658 if (z->capacity < f->zbd_info->zone_size && !td_random(td) &&
1659 ddir == DDIR_WRITE &&
1660 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1662 "%s: Jump from zone capacity limit to zone end:"
1663 " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n",
1664 f->file_name, f->last_pos[ddir],
1665 zbd_zone_end(z), zone_idx, z->capacity);
1666 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1667 f->last_pos[ddir] = zbd_zone_end(z);
1671 * zone_skip is valid only for sequential workloads.
1673 if (td_random(td) || !td->o.zone_skip)
1677 * It is time to switch to a new zone if:
1678 * - zone_bytes == zone_size bytes have already been accessed
1679 * - The last position reached the end of the current zone.
1680 * - For reads with td->o.read_beyond_wp == false, the last position
1681 * reached the zone write pointer.
1683 if (td->zone_bytes >= td->o.zone_size ||
1684 f->last_pos[ddir] >= zbd_zone_end(z) ||
1685 (ddir == DDIR_READ &&
1686 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1691 f->file_offset += td->o.zone_size + td->o.zone_skip;
1694 * Wrap from the beginning, if we exceed the file size
1696 if (f->file_offset >= f->real_file_size)
1697 f->file_offset = get_start_offset(td, f);
1699 f->last_pos[ddir] = f->file_offset;
1700 td->io_skip_bytes += td->o.zone_skip;
1705 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1707 * @td: FIO thread data.
1708 * @io_u: FIO I/O unit.
1709 * @ddir: I/O direction before adjustment.
1711 * Return adjusted I/O direction.
1713 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1717 * In case read direction is chosen for the first random I/O, fio with
1718 * zonemode=zbd stops because no data can be read from zoned block
1719 * devices with all empty zones. Overwrite the first I/O direction as
1720 * write to make sure data to read exists.
1722 assert(io_u->file->zbd_info);
1723 if (ddir != DDIR_READ || !td_rw(td))
1726 if (io_u->file->zbd_info->sectors_with_data ||
1727 td->o.read_beyond_wp)
1734 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1735 * @td: FIO thread data.
1736 * @io_u: FIO I/O unit.
1738 * Locking strategy: returns with z->mutex locked if and only if z refers
1739 * to a sequential zone and if io_u_accept is returned. z is the zone that
1740 * corresponds to io_u->offset at the end of this function.
1742 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1744 struct fio_file *f = io_u->file;
1745 struct zoned_block_device_info *zbdi = f->zbd_info;
1746 uint32_t zone_idx_b;
1747 struct fio_zone_info *zb, *zl, *orig_zb;
1748 uint32_t orig_len = io_u->buflen;
1749 uint64_t min_bs = td->o.min_bs[io_u->ddir];
1755 assert(is_valid_offset(f, io_u->offset));
1756 assert(io_u->buflen);
1757 zone_idx_b = zbd_zone_idx(f, io_u->offset);
1758 zb = get_zone(f, zone_idx_b);
1762 /* Accept non-write I/Os for conventional zones. */
1763 if (io_u->ddir != DDIR_WRITE)
1766 * Make sure that writes to conventional zones
1767 * don't cross over to any sequential zones.
1769 if (!(zb + 1)->has_wp ||
1770 io_u->offset + io_u->buflen <= (zb + 1)->start)
1773 if (io_u->offset + min_bs > (zb + 1)->start) {
1775 "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n",
1776 f->file_name, io_u->offset,
1777 min_bs, (zb + 1)->start);
1778 io_u->offset = zb->start + (zb + 1)->start - io_u->offset;
1779 new_len = min(io_u->buflen, (zb + 1)->start - io_u->offset);
1781 new_len = (zb + 1)->start - io_u->offset;
1783 io_u->buflen = new_len / min_bs * min_bs;
1788 * Accept the I/O offset for reads if reading beyond the write pointer
1791 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1792 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1795 zbd_check_swd(td, f);
1797 zone_lock(td, f, zb);
1799 switch (io_u->ddir) {
1801 if (td->runstate == TD_VERIFYING && td_write(td)) {
1802 zb = zbd_replay_write_order(td, io_u, zb);
1806 * Check that there is enough written data in the zone to do an
1807 * I/O of at least min_bs B. If there isn't, find a new zone for
1810 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1811 zb->wp - zb->start : 0;
1812 if (range < min_bs ||
1813 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1815 zl = get_zone(f, f->max_zone);
1816 zb = zbd_find_zone(td, io_u, min_bs, zb, zl);
1819 "%s: zbd_find_zone(%lld, %llu) failed\n",
1820 f->file_name, io_u->offset,
1825 * zbd_find_zone() returned a zone with a range of at
1828 range = zb->wp - zb->start;
1829 assert(range >= min_bs);
1832 io_u->offset = zb->start;
1835 * Make sure the I/O is within the zone valid data range while
1836 * maximizing the I/O size and preserving randomness.
1838 if (range <= io_u->buflen)
1839 io_u->offset = zb->start;
1840 else if (td_random(td))
1841 io_u->offset = zb->start +
1842 ((io_u->offset - orig_zb->start) %
1843 (range - io_u->buflen)) / min_bs * min_bs;
1845 * When zbd_find_zone() returns a conventional zone,
1846 * we can simply accept the new i/o offset here.
1851 * Make sure the I/O does not cross over the zone wp position.
1853 new_len = min((unsigned long long)io_u->buflen,
1854 (unsigned long long)(zb->wp - io_u->offset));
1855 new_len = new_len / min_bs * min_bs;
1856 if (new_len < io_u->buflen) {
1857 io_u->buflen = new_len;
1858 dprint(FD_IO, "Changed length from %u into %llu\n",
1859 orig_len, io_u->buflen);
1861 assert(zb->start <= io_u->offset);
1862 assert(io_u->offset + io_u->buflen <= zb->wp);
1865 if (io_u->buflen > zbdi->zone_size) {
1866 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
1868 "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n",
1869 f->file_name, io_u->buflen, zbdi->zone_size);
1872 if (!zbd_open_zone(td, f, zone_idx_b)) {
1874 zb = zbd_convert_to_open_zone(td, io_u);
1876 dprint(FD_IO, "%s: can't convert to open zone",
1881 /* Check whether the zone reset threshold has been exceeded */
1882 if (td->o.zrf.u.f) {
1883 if (zbdi->wp_sectors_with_data >=
1884 f->io_size * td->o.zrt.u.f &&
1885 zbd_dec_and_reset_write_cnt(td, f)) {
1889 /* Reset the zone pointer if necessary */
1890 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
1891 assert(td->o.verify == VERIFY_NONE);
1893 * Since previous write requests may have been submitted
1894 * asynchronously and since we will submit the zone
1895 * reset synchronously, wait until previously submitted
1896 * write requests have completed before issuing a
1901 if (zbd_reset_zone(td, f, zb) < 0)
1904 if (zb->capacity < min_bs) {
1905 td_verror(td, EINVAL, "ZCAP is less min_bs");
1906 log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n",
1907 zb->capacity, min_bs);
1911 /* Make writes occur at the write pointer */
1912 assert(!zbd_zone_full(f, zb, min_bs));
1913 io_u->offset = zb->wp;
1914 if (!is_valid_offset(f, io_u->offset)) {
1915 td_verror(td, EINVAL, "invalid WP value");
1916 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
1917 f->file_name, io_u->offset);
1921 * Make sure that the buflen is a multiple of the minimal
1922 * block size. Give up if shrinking would make the request too
1925 new_len = min((unsigned long long)io_u->buflen,
1926 zbd_zone_capacity_end(zb) - io_u->offset);
1927 new_len = new_len / min_bs * min_bs;
1928 if (new_len == io_u->buflen)
1930 if (new_len >= min_bs) {
1931 io_u->buflen = new_len;
1932 dprint(FD_IO, "Changed length from %u into %llu\n",
1933 orig_len, io_u->buflen);
1936 td_verror(td, EIO, "zone remainder too small");
1937 log_err("zone remainder %lld smaller than min block size %"PRIu64"\n",
1938 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
1941 /* Check random trim targets a non-empty zone */
1942 if (!td_random(td) || zb->wp > zb->start)
1945 /* Find out a non-empty zone to trim */
1947 zl = get_zone(f, f->max_zone);
1948 zb = zbd_find_zone(td, io_u, 1, zb, zl);
1950 io_u->offset = zb->start;
1951 dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n",
1952 f->file_name, io_u->offset);
1959 case DDIR_SYNC_FILE_RANGE:
1970 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
1971 assert(!io_u->zbd_queue_io);
1972 assert(!io_u->zbd_put_io);
1973 io_u->zbd_queue_io = zbd_queue_io;
1974 io_u->zbd_put_io = zbd_put_io;
1976 * Since we return with the zone lock still held,
1977 * add an annotation to let Coverity know that it
1980 /* coverity[missing_unlock] */
1984 if (zb && zb->has_wp)
1989 /* Return a string with ZBD statistics */
1990 char *zbd_write_status(const struct thread_stat *ts)
1994 if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0)
2000 * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim
2002 * @td: FIO thread data.
2003 * @io_u: FIO I/O unit.
2005 * It is assumed that z->mutex is already locked.
2006 * Return io_u_completed when reset zone succeeds. Return 0 when the target zone
2007 * does not have write pointer. On error, return negative errno.
2009 int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u)
2011 struct fio_file *f = io_u->file;
2012 struct fio_zone_info *z;
2016 zone_idx = zbd_zone_idx(f, io_u->offset);
2017 z = get_zone(f, zone_idx);
2022 if (io_u->offset != z->start) {
2023 log_err("Trim offset not at zone start (%lld)\n", io_u->offset);
2027 ret = zbd_reset_zone((struct thread_data *)td, f, z);
2031 return io_u_completed;