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",
256 f->file_name, zbd_zone_nr(f, z));
258 switch (f->zbd_info->model) {
260 case ZBD_HOST_MANAGED:
261 ret = zbd_reset_wp(td, f, offset, length);
269 pthread_mutex_lock(&f->zbd_info->mutex);
270 f->zbd_info->sectors_with_data -= data_in_zone;
271 f->zbd_info->wp_sectors_with_data -= data_in_zone;
272 pthread_mutex_unlock(&f->zbd_info->mutex);
277 td->ts.nr_zone_resets++;
283 * zbd_close_zone - Remove a zone from the open zones array.
284 * @td: FIO thread data.
285 * @f: FIO file associated with the disk for which to reset a write pointer.
286 * @zone_idx: Index of the zone to remove.
288 * The caller must hold f->zbd_info->mutex.
290 static void zbd_close_zone(struct thread_data *td, const struct fio_file *f,
291 unsigned int zone_idx)
293 uint32_t open_zone_idx = 0;
295 for (; open_zone_idx < f->zbd_info->num_open_zones; open_zone_idx++) {
296 if (f->zbd_info->open_zones[open_zone_idx] == zone_idx)
299 if (open_zone_idx == f->zbd_info->num_open_zones)
302 dprint(FD_ZBD, "%s: closing zone %d\n",
303 f->file_name, zone_idx);
305 memmove(f->zbd_info->open_zones + open_zone_idx,
306 f->zbd_info->open_zones + open_zone_idx + 1,
307 (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) *
308 sizeof(f->zbd_info->open_zones[0]));
310 f->zbd_info->num_open_zones--;
311 td->num_open_zones--;
312 get_zone(f, zone_idx)->open = 0;
316 * zbd_reset_zones - Reset a range of zones.
317 * @td: fio thread data.
318 * @f: fio file for which to reset zones
319 * @zb: first zone to reset.
320 * @ze: first zone not to reset.
322 * Returns 0 upon success and 1 upon failure.
324 static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
325 struct fio_zone_info *const zb,
326 struct fio_zone_info *const ze)
328 struct fio_zone_info *z;
329 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
334 dprint(FD_ZBD, "%s: examining zones %u .. %u\n",
335 f->file_name, zbd_zone_nr(f, zb), zbd_zone_nr(f, ze));
337 for (z = zb; z < ze; z++) {
338 uint32_t nz = zbd_zone_nr(f, z);
344 pthread_mutex_lock(&f->zbd_info->mutex);
345 zbd_close_zone(td, f, nz);
346 pthread_mutex_unlock(&f->zbd_info->mutex);
348 if (z->wp != z->start) {
349 dprint(FD_ZBD, "%s: resetting zone %u\n",
350 f->file_name, zbd_zone_nr(f, z));
351 if (zbd_reset_zone(td, f, z) < 0)
362 * zbd_get_max_open_zones - Get the maximum number of open zones
363 * @td: FIO thread data
364 * @f: FIO file for which to get max open zones
365 * @max_open_zones: Upon success, result will be stored here.
367 * A @max_open_zones value set to zero means no limit.
369 * Returns 0 upon success and a negative error code upon failure.
371 static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f,
372 unsigned int *max_open_zones)
376 if (td->io_ops && td->io_ops->get_max_open_zones)
377 ret = td->io_ops->get_max_open_zones(td, f, max_open_zones);
379 ret = blkzoned_get_max_open_zones(td, f, max_open_zones);
381 td_verror(td, errno, "get max open zones failed");
382 log_err("%s: get max open zones failed (%d).\n",
383 f->file_name, errno);
390 * zbd_open_zone - Add a zone to the array of open zones.
391 * @td: fio thread data.
392 * @f: fio file that has the open zones to add.
393 * @zone_idx: Index of the zone to add.
395 * Open a ZBD zone if it is not already open. Returns true if either the zone
396 * was already open or if the zone was successfully added to the array of open
397 * zones without exceeding the maximum number of open zones. Returns false if
398 * the zone was not already open and opening the zone would cause the zone limit
401 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
404 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
405 struct zoned_block_device_info *zbdi = f->zbd_info;
406 struct fio_zone_info *z = get_zone(f, zone_idx);
409 if (z->cond == ZBD_ZONE_COND_OFFLINE)
413 * Skip full zones with data verification enabled because resetting a
414 * zone causes data loss and hence causes verification to fail.
416 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
420 * zbdi->max_open_zones == 0 means that there is no limit on the maximum
421 * number of open zones. In this case, do no track open zones in
422 * zbdi->open_zones array.
424 if (!zbdi->max_open_zones)
427 pthread_mutex_lock(&zbdi->mutex);
431 * If the zone is going to be completely filled by writes
432 * already in-flight, handle it as a full zone instead of an
435 if (z->wp >= zbd_zone_capacity_end(z))
441 /* Zero means no limit */
442 if (td->o.job_max_open_zones > 0 &&
443 td->num_open_zones >= td->o.job_max_open_zones)
445 if (zbdi->num_open_zones >= zbdi->max_open_zones)
448 dprint(FD_ZBD, "%s: opening zone %d\n",
449 f->file_name, zone_idx);
451 zbdi->open_zones[zbdi->num_open_zones++] = zone_idx;
452 td->num_open_zones++;
457 pthread_mutex_unlock(&zbdi->mutex);
461 /* Verify whether direct I/O is used for all host-managed zoned drives. */
462 static bool zbd_using_direct_io(void)
464 struct thread_data *td;
469 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
471 for_each_file(td, f, j) {
473 f->zbd_info->model == ZBD_HOST_MANAGED)
481 /* Whether or not the I/O range for f includes one or more sequential zones */
482 static bool zbd_is_seq_job(struct fio_file *f)
484 uint32_t zone_idx, zone_idx_b, zone_idx_e;
491 zone_idx_b = zbd_zone_idx(f, f->file_offset);
492 zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1);
493 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
494 if (get_zone(f, zone_idx)->has_wp)
501 * Verify whether the file offset and size parameters are aligned with zone
502 * boundaries. If the file offset is not aligned, align it down to the start of
503 * the zone containing the start offset and align up the file io_size parameter.
505 static bool zbd_zone_align_file_sizes(struct thread_data *td,
508 const struct fio_zone_info *z;
509 uint64_t new_offset, new_end;
514 if (f->file_offset >= f->real_file_size)
516 if (!zbd_is_seq_job(f))
519 if (!td->o.zone_size) {
520 td->o.zone_size = f->zbd_info->zone_size;
521 if (!td->o.zone_size) {
522 log_err("%s: invalid 0 zone size\n",
526 } else if (td->o.zone_size != f->zbd_info->zone_size) {
527 log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n",
528 f->file_name, td->o.zone_size,
529 f->zbd_info->zone_size);
533 if (td->o.zone_skip % td->o.zone_size) {
534 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
535 f->file_name, td->o.zone_skip,
540 zone_idx = zbd_zone_idx(f, f->file_offset);
541 z = get_zone(f, zone_idx);
542 if ((f->file_offset != z->start) &&
543 (td->o.td_ddir != TD_DDIR_READ)) {
544 new_offset = zbd_zone_end(z);
545 if (new_offset >= f->file_offset + f->io_size) {
546 log_info("%s: io_size must be at least one zone\n",
550 log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n",
551 f->file_name, f->file_offset,
553 f->io_size -= (new_offset - f->file_offset);
554 f->file_offset = new_offset;
557 zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size);
558 z = get_zone(f, zone_idx);
560 if ((td->o.td_ddir != TD_DDIR_READ) &&
561 (f->file_offset + f->io_size != new_end)) {
562 if (new_end <= f->file_offset) {
563 log_info("%s: io_size must be at least one zone\n",
567 log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n",
568 f->file_name, f->io_size,
569 new_end - f->file_offset);
570 f->io_size = new_end - f->file_offset;
577 * Verify whether offset and size parameters are aligned with zone boundaries.
579 static bool zbd_verify_sizes(void)
581 struct thread_data *td;
586 for_each_file(td, f, j) {
587 if (!zbd_zone_align_file_sizes(td, f))
595 static bool zbd_verify_bs(void)
597 struct thread_data *td;
603 (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] ||
604 td->o.bssplit_nr[DDIR_TRIM])) {
605 log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n");
608 for_each_file(td, f, j) {
614 zone_size = f->zbd_info->zone_size;
615 if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) {
616 log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n",
617 f->file_name, td->o.bs[DDIR_TRIM],
621 for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) {
622 if (td->o.verify != VERIFY_NONE &&
623 zone_size % td->o.bs[k] != 0) {
624 log_info("%s: block size %llu is not a divisor of the zone size %"PRIu64"\n",
625 f->file_name, td->o.bs[k],
635 static int ilog2(uint64_t i)
647 * Initialize f->zbd_info for devices that are not zoned block devices. This
648 * allows to execute a ZBD workload against a non-ZBD device.
650 static int init_zone_info(struct thread_data *td, struct fio_file *f)
653 struct fio_zone_info *p;
654 uint64_t zone_size = td->o.zone_size;
655 uint64_t zone_capacity = td->o.zone_capacity;
656 struct zoned_block_device_info *zbd_info = NULL;
659 if (zone_size == 0) {
660 log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n",
665 if (zone_size < 512) {
666 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
671 if (zone_capacity == 0)
672 zone_capacity = zone_size;
674 if (zone_capacity > zone_size) {
675 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
676 f->file_name, td->o.zone_capacity, td->o.zone_size);
680 if (f->real_file_size < zone_size) {
681 log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n",
682 f->file_name, f->real_file_size, zone_size);
686 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
687 zbd_info = scalloc(1, sizeof(*zbd_info) +
688 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
692 mutex_init_pshared(&zbd_info->mutex);
693 zbd_info->refcount = 1;
694 p = &zbd_info->zone_info[0];
695 for (i = 0; i < nr_zones; i++, p++) {
696 mutex_init_pshared_with_type(&p->mutex,
697 PTHREAD_MUTEX_RECURSIVE);
698 p->start = i * zone_size;
700 p->type = ZBD_ZONE_TYPE_SWR;
701 p->cond = ZBD_ZONE_COND_EMPTY;
702 p->capacity = zone_capacity;
706 p->start = nr_zones * zone_size;
708 f->zbd_info = zbd_info;
709 f->zbd_info->zone_size = zone_size;
710 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
711 ilog2(zone_size) : 0;
712 f->zbd_info->nr_zones = nr_zones;
717 * Maximum number of zones to report in one operation.
719 #define ZBD_REPORT_MAX_ZONES 8192U
722 * Parse the device zone report and store it in f->zbd_info. Must be called
723 * only for devices that are zoned, namely those with a model != ZBD_NONE.
725 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
728 struct zbd_zone *zones, *z;
729 struct fio_zone_info *p;
730 uint64_t zone_size, offset;
731 struct zoned_block_device_info *zbd_info = NULL;
732 int i, j, ret = -ENOMEM;
734 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
738 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
741 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
746 zone_size = zones[0].len;
747 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
749 if (td->o.zone_size == 0) {
750 td->o.zone_size = zone_size;
751 } else if (td->o.zone_size != zone_size) {
752 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
753 f->file_name, td->o.zone_size, zone_size);
758 dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n",
759 f->file_name, nr_zones, zone_size / 1024);
761 zbd_info = scalloc(1, sizeof(*zbd_info) +
762 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
765 mutex_init_pshared(&zbd_info->mutex);
766 zbd_info->refcount = 1;
767 p = &zbd_info->zone_info[0];
768 for (offset = 0, j = 0; j < nr_zones;) {
770 for (i = 0; i < nrz; i++, j++, z++, p++) {
771 mutex_init_pshared_with_type(&p->mutex,
772 PTHREAD_MUTEX_RECURSIVE);
774 p->capacity = z->capacity;
777 case ZBD_ZONE_COND_NOT_WP:
778 case ZBD_ZONE_COND_FULL:
779 p->wp = p->start + p->capacity;
782 assert(z->start <= z->wp);
783 assert(z->wp <= z->start + zone_size);
789 case ZBD_ZONE_TYPE_SWR:
798 if (j > 0 && p->start != p[-1].start + zone_size) {
799 log_info("%s: invalid zone data\n",
806 offset = z->start + z->len;
810 nrz = zbd_report_zones(td, f, offset, zones,
811 min((uint32_t)(nr_zones - j),
812 ZBD_REPORT_MAX_ZONES));
815 log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n",
816 offset, f->file_name, -ret);
822 zbd_info->zone_info[nr_zones].start = offset;
824 f->zbd_info = zbd_info;
825 f->zbd_info->zone_size = zone_size;
826 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
827 ilog2(zone_size) : 0;
828 f->zbd_info->nr_zones = nr_zones;
838 static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f)
840 struct zoned_block_device_info *zbd = f->zbd_info;
841 unsigned int max_open_zones;
844 if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) {
845 /* Only host-managed devices have a max open limit */
846 zbd->max_open_zones = td->o.max_open_zones;
850 /* If host-managed, get the max open limit */
851 ret = zbd_get_max_open_zones(td, f, &max_open_zones);
855 if (!max_open_zones) {
856 /* No device limit */
857 zbd->max_open_zones = td->o.max_open_zones;
858 } else if (!td->o.max_open_zones) {
859 /* No user limit. Set limit to device limit */
860 zbd->max_open_zones = max_open_zones;
861 } else if (td->o.max_open_zones <= max_open_zones) {
862 /* Both user limit and dev limit. User limit not too large */
863 zbd->max_open_zones = td->o.max_open_zones;
865 /* Both user limit and dev limit. User limit too large */
866 td_verror(td, EINVAL,
867 "Specified --max_open_zones is too large");
868 log_err("Specified --max_open_zones (%d) is larger than max (%u)\n",
869 td->o.max_open_zones, max_open_zones);
874 /* Ensure that the limit is not larger than FIO's internal limit */
875 if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) {
876 td_verror(td, EINVAL, "'max_open_zones' value is too large");
877 log_err("'max_open_zones' value is larger than %u\n",
882 dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n",
883 f->file_name, zbd->max_open_zones);
889 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
891 * Returns 0 upon success and a negative error code upon failure.
893 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
895 enum zbd_zoned_model zbd_model;
898 assert(td->o.zone_mode == ZONE_MODE_ZBD);
900 ret = zbd_get_zoned_model(td, f, &zbd_model);
906 case ZBD_HOST_MANAGED:
907 ret = parse_zone_info(td, f);
912 ret = init_zone_info(td, f);
917 td_verror(td, EINVAL, "Unsupported zoned model");
918 log_err("Unsupported zoned model\n");
923 f->zbd_info->model = zbd_model;
925 ret = zbd_set_max_open_zones(td, f);
927 zbd_free_zone_info(f);
934 void zbd_free_zone_info(struct fio_file *f)
940 pthread_mutex_lock(&f->zbd_info->mutex);
941 refcount = --f->zbd_info->refcount;
942 pthread_mutex_unlock(&f->zbd_info->mutex);
944 assert((int32_t)refcount >= 0);
951 * Initialize f->zbd_info.
953 * Returns 0 upon success and a negative error code upon failure.
955 * Note: this function can only work correctly if it is called before the first
958 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
960 struct thread_data *td2;
964 for_each_td(td2, i) {
965 for_each_file(td2, f2, j) {
966 if (td2 == td && f2 == file)
969 strcmp(f2->file_name, file->file_name) != 0)
971 file->zbd_info = f2->zbd_info;
972 file->zbd_info->refcount++;
977 ret = zbd_create_zone_info(td, file);
979 td_verror(td, -ret, "zbd_create_zone_info() failed");
984 int zbd_init_files(struct thread_data *td)
989 for_each_file(td, f, i) {
990 if (zbd_init_zone_info(td, f))
997 void zbd_recalc_options_with_zone_granularity(struct thread_data *td)
1002 for_each_file(td, f, i) {
1003 struct zoned_block_device_info *zbd = f->zbd_info;
1006 /* zonemode=strided doesn't get per-file zone size. */
1007 zone_size = zbd ? zbd->zone_size : td->o.zone_size;
1011 if (td->o.size_nz > 0)
1012 td->o.size = td->o.size_nz * zone_size;
1013 if (td->o.io_size_nz > 0)
1014 td->o.io_size = td->o.io_size_nz * zone_size;
1015 if (td->o.start_offset_nz > 0)
1016 td->o.start_offset = td->o.start_offset_nz * zone_size;
1017 if (td->o.offset_increment_nz > 0)
1018 td->o.offset_increment =
1019 td->o.offset_increment_nz * zone_size;
1020 if (td->o.zone_skip_nz > 0)
1021 td->o.zone_skip = td->o.zone_skip_nz * zone_size;
1025 int zbd_setup_files(struct thread_data *td)
1030 if (!zbd_using_direct_io()) {
1031 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
1035 if (!zbd_verify_sizes())
1038 if (!zbd_verify_bs())
1041 for_each_file(td, f, i) {
1042 struct zoned_block_device_info *zbd = f->zbd_info;
1043 struct fio_zone_info *z;
1048 f->min_zone = zbd_zone_idx(f, f->file_offset);
1049 f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size);
1052 * When all zones in the I/O range are conventional, io_size
1053 * can be smaller than zone size, making min_zone the same
1054 * as max_zone. This is why the assert below needs to be made
1057 if (zbd_is_seq_job(f))
1058 assert(f->min_zone < f->max_zone);
1060 if (td->o.max_open_zones > 0 &&
1061 zbd->max_open_zones != td->o.max_open_zones) {
1062 log_err("Different 'max_open_zones' values\n");
1067 * The per job max open zones limit cannot be used without a
1068 * global max open zones limit. (As the tracking of open zones
1069 * is disabled when there is no global max open zones limit.)
1071 if (td->o.job_max_open_zones && !zbd->max_open_zones) {
1072 log_err("'job_max_open_zones' cannot be used without a global open zones limit\n");
1077 * zbd->max_open_zones is the global limit shared for all jobs
1078 * that target the same zoned block device. Force sync the per
1079 * thread global limit with the actual global limit. (The real
1080 * per thread/job limit is stored in td->o.job_max_open_zones).
1082 td->o.max_open_zones = zbd->max_open_zones;
1084 for (zi = f->min_zone; zi < f->max_zone; zi++) {
1085 z = &zbd->zone_info[zi];
1086 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
1087 z->cond != ZBD_ZONE_COND_EXP_OPEN)
1089 if (zbd_open_zone(td, f, zi))
1092 * If the number of open zones exceeds specified limits,
1093 * reset all extra open zones.
1095 if (zbd_reset_zone(td, f, z) < 0) {
1096 log_err("Failed to reest zone %d\n", zi);
1106 * Reset zbd_info.write_cnt, the counter that counts down towards the next
1109 static void _zbd_reset_write_cnt(const struct thread_data *td,
1110 const struct fio_file *f)
1112 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
1114 f->zbd_info->write_cnt = td->o.zrf.u.f ?
1115 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
1118 static void zbd_reset_write_cnt(const struct thread_data *td,
1119 const struct fio_file *f)
1121 pthread_mutex_lock(&f->zbd_info->mutex);
1122 _zbd_reset_write_cnt(td, f);
1123 pthread_mutex_unlock(&f->zbd_info->mutex);
1126 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
1127 const struct fio_file *f)
1129 uint32_t write_cnt = 0;
1131 pthread_mutex_lock(&f->zbd_info->mutex);
1132 assert(f->zbd_info->write_cnt);
1133 if (f->zbd_info->write_cnt)
1134 write_cnt = --f->zbd_info->write_cnt;
1136 _zbd_reset_write_cnt(td, f);
1137 pthread_mutex_unlock(&f->zbd_info->mutex);
1139 return write_cnt == 0;
1147 /* Calculate the number of sectors with data (swd) and perform action 'a' */
1148 static uint64_t zbd_process_swd(struct thread_data *td,
1149 const struct fio_file *f, enum swd_action a)
1151 struct fio_zone_info *zb, *ze, *z;
1153 uint64_t wp_swd = 0;
1155 zb = get_zone(f, f->min_zone);
1156 ze = get_zone(f, f->max_zone);
1157 for (z = zb; z < ze; z++) {
1159 zone_lock(td, f, z);
1160 wp_swd += z->wp - z->start;
1162 swd += z->wp - z->start;
1165 pthread_mutex_lock(&f->zbd_info->mutex);
1168 assert(f->zbd_info->sectors_with_data == swd);
1169 assert(f->zbd_info->wp_sectors_with_data == wp_swd);
1172 f->zbd_info->sectors_with_data = swd;
1173 f->zbd_info->wp_sectors_with_data = wp_swd;
1176 pthread_mutex_unlock(&f->zbd_info->mutex);
1178 for (z = zb; z < ze; z++)
1186 * The swd check is useful for debugging but takes too much time to leave
1187 * it enabled all the time. Hence it is disabled by default.
1189 static const bool enable_check_swd = false;
1191 /* Check whether the values of zbd_info.*sectors_with_data are correct. */
1192 static void zbd_check_swd(struct thread_data *td, const struct fio_file *f)
1194 if (!enable_check_swd)
1197 zbd_process_swd(td, f, CHECK_SWD);
1200 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
1202 struct fio_zone_info *zb, *ze;
1205 if (!f->zbd_info || !td_write(td))
1208 zb = get_zone(f, f->min_zone);
1209 ze = get_zone(f, f->max_zone);
1210 swd = zbd_process_swd(td, f, SET_SWD);
1212 dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n",
1213 __func__, f->file_name, swd);
1216 * If data verification is enabled reset the affected zones before
1217 * writing any data to avoid that a zone reset has to be issued while
1218 * writing data, which causes data loss.
1220 if (td->o.verify != VERIFY_NONE && td->runstate != TD_VERIFYING)
1221 zbd_reset_zones(td, f, zb, ze);
1222 zbd_reset_write_cnt(td, f);
1225 /* Return random zone index for one of the open zones. */
1226 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1227 const struct io_u *io_u)
1229 return (io_u->offset - f->file_offset) *
1230 f->zbd_info->num_open_zones / f->io_size;
1233 static bool any_io_in_flight(void)
1235 struct thread_data *td;
1238 for_each_td(td, i) {
1239 if (td->io_u_in_flight)
1247 * Modify the offset of an I/O unit that does not refer to an open zone such
1248 * that it refers to an open zone. Close an open zone and open a new zone if
1249 * necessary. The open zone is searched across sequential zones.
1250 * This algorithm can only work correctly if all write pointers are
1251 * a multiple of the fio block size. The caller must neither hold z->mutex
1252 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1254 static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
1257 const uint64_t min_bs = td->o.min_bs[io_u->ddir];
1258 struct fio_file *f = io_u->file;
1259 struct zoned_block_device_info *zbdi = f->zbd_info;
1260 struct fio_zone_info *z;
1261 unsigned int open_zone_idx = -1;
1262 uint32_t zone_idx, new_zone_idx;
1264 bool wait_zone_close;
1266 bool should_retry = true;
1268 assert(is_valid_offset(f, io_u->offset));
1270 if (zbdi->max_open_zones || td->o.job_max_open_zones) {
1272 * This statement accesses zbdi->open_zones[] on purpose
1275 zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)];
1277 zone_idx = zbd_zone_idx(f, io_u->offset);
1279 if (zone_idx < f->min_zone)
1280 zone_idx = f->min_zone;
1281 else if (zone_idx >= f->max_zone)
1282 zone_idx = f->max_zone - 1;
1285 "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1286 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1289 * Since z->mutex is the outer lock and zbdi->mutex the inner
1290 * lock it can happen that the state of the zone with index zone_idx
1291 * has changed after 'z' has been assigned and before zbdi->mutex
1292 * has been obtained. Hence the loop.
1297 z = get_zone(f, zone_idx);
1299 zone_lock(td, f, z);
1301 pthread_mutex_lock(&zbdi->mutex);
1304 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1305 zbdi->max_open_zones == 0 &&
1306 td->o.job_max_open_zones == 0)
1308 if (zbdi->num_open_zones == 0) {
1309 dprint(FD_ZBD, "%s(%s): no zones are open\n",
1310 __func__, f->file_name);
1311 goto open_other_zone;
1316 * List of opened zones is per-device, shared across all
1317 * threads. Start with quasi-random candidate zone. Ignore
1318 * zones which don't belong to thread's offset/size area.
1320 open_zone_idx = pick_random_zone_idx(f, io_u);
1321 assert(!open_zone_idx ||
1322 open_zone_idx < zbdi->num_open_zones);
1323 tmp_idx = open_zone_idx;
1325 for (i = 0; i < zbdi->num_open_zones; i++) {
1328 if (tmp_idx >= zbdi->num_open_zones)
1330 tmpz = zbdi->open_zones[tmp_idx];
1331 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1332 open_zone_idx = tmp_idx;
1333 goto found_candidate_zone;
1339 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1340 __func__, f->file_name);
1342 pthread_mutex_unlock(&zbdi->mutex);
1349 found_candidate_zone:
1350 new_zone_idx = zbdi->open_zones[open_zone_idx];
1351 if (new_zone_idx == zone_idx)
1353 zone_idx = new_zone_idx;
1355 pthread_mutex_unlock(&zbdi->mutex);
1361 /* Both z->mutex and zbdi->mutex are held. */
1364 if (z->wp + min_bs <= zbd_zone_capacity_end(z)) {
1365 pthread_mutex_unlock(&zbdi->mutex);
1370 /* Check if number of open zones reaches one of limits. */
1372 zbdi->num_open_zones == f->max_zone - f->min_zone ||
1373 (zbdi->max_open_zones &&
1374 zbdi->num_open_zones == zbdi->max_open_zones) ||
1375 (td->o.job_max_open_zones &&
1376 td->num_open_zones == td->o.job_max_open_zones);
1378 pthread_mutex_unlock(&zbdi->mutex);
1380 /* Only z->mutex is held. */
1383 * When number of open zones reaches to one of limits, wait for
1384 * zone close before opening a new zone.
1386 if (wait_zone_close) {
1388 "%s(%s): quiesce to allow open zones to close\n",
1389 __func__, f->file_name);
1394 /* Zone 'z' is full, so try to open a new zone. */
1395 for (i = f->io_size / zbdi->zone_size; i > 0; i--) {
1400 if (!is_valid_offset(f, z->start)) {
1402 zone_idx = f->min_zone;
1403 z = get_zone(f, zone_idx);
1405 assert(is_valid_offset(f, z->start));
1408 zone_lock(td, f, z);
1411 if (zbd_open_zone(td, f, zone_idx))
1415 /* Only z->mutex is held. */
1417 /* Check whether the write fits in any of the already opened zones. */
1418 pthread_mutex_lock(&zbdi->mutex);
1419 for (i = 0; i < zbdi->num_open_zones; i++) {
1420 zone_idx = zbdi->open_zones[i];
1421 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1423 pthread_mutex_unlock(&zbdi->mutex);
1426 z = get_zone(f, zone_idx);
1428 zone_lock(td, f, z);
1429 if (z->wp + min_bs <= zbd_zone_capacity_end(z))
1431 pthread_mutex_lock(&zbdi->mutex);
1435 * When any I/O is in-flight or when all I/Os in-flight get completed,
1436 * the I/Os might have closed zones then retry the steps to open a zone.
1437 * Before retry, call io_u_quiesce() to complete in-flight writes.
1439 in_flight = any_io_in_flight();
1440 if (in_flight || should_retry) {
1442 "%s(%s): wait zone close and retry open zones\n",
1443 __func__, f->file_name);
1444 pthread_mutex_unlock(&zbdi->mutex);
1447 zone_lock(td, f, z);
1448 should_retry = in_flight;
1452 pthread_mutex_unlock(&zbdi->mutex);
1456 dprint(FD_ZBD, "%s(%s): did not open another zone\n",
1457 __func__, f->file_name);
1462 dprint(FD_ZBD, "%s(%s): returning zone %d\n",
1463 __func__, f->file_name, zone_idx);
1465 io_u->offset = z->start;
1467 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1472 /* The caller must hold z->mutex. */
1473 static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td,
1475 struct fio_zone_info *z)
1477 const struct fio_file *f = io_u->file;
1478 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
1480 if (!zbd_open_zone(td, f, zbd_zone_nr(f, z))) {
1482 z = zbd_convert_to_open_zone(td, io_u);
1486 if (z->verify_block * min_bs >= z->capacity) {
1487 log_err("%s: %d * %"PRIu64" >= %"PRIu64"\n",
1488 f->file_name, z->verify_block, min_bs, z->capacity);
1490 * If the assertion below fails during a test run, adding
1491 * "--experimental_verify=1" to the command line may help.
1496 io_u->offset = z->start + z->verify_block * min_bs;
1497 if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1498 log_err("%s: %llu + %llu >= %"PRIu64"\n",
1499 f->file_name, io_u->offset, io_u->buflen,
1500 zbd_zone_capacity_end(z));
1503 z->verify_block += io_u->buflen / min_bs;
1509 * Find another zone which has @min_bytes of readable data. Search in zones
1510 * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf.
1512 * Either returns NULL or returns a zone pointer. When the zone has write
1513 * pointer, hold the mutex for the zone.
1515 static struct fio_zone_info *
1516 zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes,
1517 struct fio_zone_info *zb, struct fio_zone_info *zl)
1519 struct fio_file *f = io_u->file;
1520 struct fio_zone_info *z1, *z2;
1521 const struct fio_zone_info *const zf = get_zone(f, f->min_zone);
1524 * Skip to the next non-empty zone in case of sequential I/O and to
1525 * the nearest non-empty zone in case of random I/O.
1527 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1528 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1530 zone_lock(td, f, z1);
1531 if (z1->start + min_bytes <= z1->wp)
1535 } else if (!td_random(td)) {
1539 if (td_random(td) && z2 >= zf &&
1540 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1542 zone_lock(td, f, z2);
1543 if (z2->start + min_bytes <= z2->wp)
1551 "%s: no zone has %"PRIu64" bytes of readable data\n",
1552 f->file_name, min_bytes);
1558 * zbd_end_zone_io - update zone status at command completion
1560 * @z: zone info pointer
1562 * If the write command made the zone full, close it.
1564 * The caller must hold z->mutex.
1566 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1567 struct fio_zone_info *z)
1569 const struct fio_file *f = io_u->file;
1571 if (io_u->ddir == DDIR_WRITE &&
1572 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1573 pthread_mutex_lock(&f->zbd_info->mutex);
1574 zbd_close_zone(td, f, zbd_zone_nr(f, z));
1575 pthread_mutex_unlock(&f->zbd_info->mutex);
1580 * zbd_queue_io - update the write pointer of a sequential zone
1582 * @success: Whether or not the I/O unit has been queued successfully
1583 * @q: queueing status (busy, completed or queued).
1585 * For write and trim operations, update the write pointer of the I/O unit
1588 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1591 const struct fio_file *f = io_u->file;
1592 struct zoned_block_device_info *zbd_info = f->zbd_info;
1593 struct fio_zone_info *z;
1599 zone_idx = zbd_zone_idx(f, io_u->offset);
1600 assert(zone_idx < zbd_info->nr_zones);
1601 z = get_zone(f, zone_idx);
1609 "%s: queued I/O (%lld, %llu) for zone %u\n",
1610 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1612 switch (io_u->ddir) {
1614 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1615 zbd_zone_capacity_end(z));
1618 * z->wp > zone_end means that one or more I/O errors
1621 pthread_mutex_lock(&zbd_info->mutex);
1622 if (z->wp <= zone_end) {
1623 zbd_info->sectors_with_data += zone_end - z->wp;
1624 zbd_info->wp_sectors_with_data += zone_end - z->wp;
1626 pthread_mutex_unlock(&zbd_info->mutex);
1633 if (q == FIO_Q_COMPLETED && !io_u->error)
1634 zbd_end_zone_io(td, io_u, z);
1637 if (!success || q != FIO_Q_QUEUED) {
1638 /* BUSY or COMPLETED: unlock the zone */
1640 io_u->zbd_put_io = NULL;
1645 * zbd_put_io - Unlock an I/O unit target zone lock
1648 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1650 const struct fio_file *f = io_u->file;
1651 struct zoned_block_device_info *zbd_info = f->zbd_info;
1652 struct fio_zone_info *z;
1657 zone_idx = zbd_zone_idx(f, io_u->offset);
1658 assert(zone_idx < zbd_info->nr_zones);
1659 z = get_zone(f, zone_idx);
1664 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1665 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1667 zbd_end_zone_io(td, io_u, z);
1670 zbd_check_swd(td, f);
1674 * Windows and MacOS do not define this.
1677 #define EREMOTEIO 121 /* POSIX value */
1680 bool zbd_unaligned_write(int error_code)
1682 switch (error_code) {
1691 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1692 * @td: FIO thread data.
1693 * @io_u: FIO I/O unit.
1695 * For sequential workloads, change the file offset to skip zoneskip bytes when
1696 * no more IO can be performed in the current zone.
1697 * - For read workloads, zoneskip is applied when the io has reached the end of
1698 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1699 * - For write workloads, zoneskip is applied when the zone is full.
1700 * This applies only to read and write operations.
1702 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1704 struct fio_file *f = io_u->file;
1705 enum fio_ddir ddir = io_u->ddir;
1706 struct fio_zone_info *z;
1709 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1710 assert(td->o.zone_size);
1711 assert(f->zbd_info);
1713 zone_idx = zbd_zone_idx(f, f->last_pos[ddir]);
1714 z = get_zone(f, zone_idx);
1717 * When the zone capacity is smaller than the zone size and the I/O is
1718 * sequential write, skip to zone end if the latest position is at the
1719 * zone capacity limit.
1721 if (z->capacity < f->zbd_info->zone_size &&
1722 !td_random(td) && ddir == DDIR_WRITE &&
1723 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1725 "%s: Jump from zone capacity limit to zone end:"
1726 " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n",
1727 f->file_name, f->last_pos[ddir],
1728 zbd_zone_end(z), zone_idx, z->capacity);
1729 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1730 f->last_pos[ddir] = zbd_zone_end(z);
1734 * zone_skip is valid only for sequential workloads.
1736 if (td_random(td) || !td->o.zone_skip)
1740 * It is time to switch to a new zone if:
1741 * - zone_bytes == zone_size bytes have already been accessed
1742 * - The last position reached the end of the current zone.
1743 * - For reads with td->o.read_beyond_wp == false, the last position
1744 * reached the zone write pointer.
1746 if (td->zone_bytes >= td->o.zone_size ||
1747 f->last_pos[ddir] >= zbd_zone_end(z) ||
1748 (ddir == DDIR_READ &&
1749 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1754 f->file_offset += td->o.zone_size + td->o.zone_skip;
1757 * Wrap from the beginning, if we exceed the file size
1759 if (f->file_offset >= f->real_file_size)
1760 f->file_offset = get_start_offset(td, f);
1762 f->last_pos[ddir] = f->file_offset;
1763 td->io_skip_bytes += td->o.zone_skip;
1768 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1770 * @td: FIO thread data.
1771 * @io_u: FIO I/O unit.
1772 * @ddir: I/O direction before adjustment.
1774 * Return adjusted I/O direction.
1776 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1780 * In case read direction is chosen for the first random I/O, fio with
1781 * zonemode=zbd stops because no data can be read from zoned block
1782 * devices with all empty zones. Overwrite the first I/O direction as
1783 * write to make sure data to read exists.
1785 assert(io_u->file->zbd_info);
1786 if (ddir != DDIR_READ || !td_rw(td))
1789 if (io_u->file->zbd_info->sectors_with_data ||
1790 td->o.read_beyond_wp)
1797 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1798 * @td: FIO thread data.
1799 * @io_u: FIO I/O unit.
1801 * Locking strategy: returns with z->mutex locked if and only if z refers
1802 * to a sequential zone and if io_u_accept is returned. z is the zone that
1803 * corresponds to io_u->offset at the end of this function.
1805 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1807 struct fio_file *f = io_u->file;
1808 struct zoned_block_device_info *zbdi = f->zbd_info;
1809 uint32_t zone_idx_b;
1810 struct fio_zone_info *zb, *zl, *orig_zb;
1811 uint32_t orig_len = io_u->buflen;
1812 uint64_t min_bs = td->o.min_bs[io_u->ddir];
1818 assert(is_valid_offset(f, io_u->offset));
1819 assert(io_u->buflen);
1821 zone_idx_b = zbd_zone_idx(f, io_u->offset);
1822 zb = get_zone(f, zone_idx_b);
1826 /* Accept non-write I/Os for conventional zones. */
1827 if (io_u->ddir != DDIR_WRITE)
1831 * Make sure that writes to conventional zones
1832 * don't cross over to any sequential zones.
1834 if (!(zb + 1)->has_wp ||
1835 io_u->offset + io_u->buflen <= (zb + 1)->start)
1838 if (io_u->offset + min_bs > (zb + 1)->start) {
1840 "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n",
1841 f->file_name, io_u->offset,
1842 min_bs, (zb + 1)->start);
1844 zb->start + (zb + 1)->start - io_u->offset;
1845 new_len = min(io_u->buflen,
1846 (zb + 1)->start - io_u->offset);
1848 new_len = (zb + 1)->start - io_u->offset;
1851 io_u->buflen = new_len / min_bs * min_bs;
1857 * Accept the I/O offset for reads if reading beyond the write pointer
1860 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1861 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1864 zbd_check_swd(td, f);
1866 zone_lock(td, f, zb);
1868 switch (io_u->ddir) {
1870 if (td->runstate == TD_VERIFYING && td_write(td)) {
1871 zb = zbd_replay_write_order(td, io_u, zb);
1876 * Check that there is enough written data in the zone to do an
1877 * I/O of at least min_bs B. If there isn't, find a new zone for
1880 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1881 zb->wp - zb->start : 0;
1882 if (range < min_bs ||
1883 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1885 zl = get_zone(f, f->max_zone);
1886 zb = zbd_find_zone(td, io_u, min_bs, zb, zl);
1889 "%s: zbd_find_zone(%lld, %llu) failed\n",
1890 f->file_name, io_u->offset,
1895 * zbd_find_zone() returned a zone with a range of at
1898 range = zb->wp - zb->start;
1899 assert(range >= min_bs);
1902 io_u->offset = zb->start;
1906 * Make sure the I/O is within the zone valid data range while
1907 * maximizing the I/O size and preserving randomness.
1909 if (range <= io_u->buflen)
1910 io_u->offset = zb->start;
1911 else if (td_random(td))
1912 io_u->offset = zb->start +
1913 ((io_u->offset - orig_zb->start) %
1914 (range - io_u->buflen)) / min_bs * min_bs;
1917 * When zbd_find_zone() returns a conventional zone,
1918 * we can simply accept the new i/o offset here.
1924 * Make sure the I/O does not cross over the zone wp position.
1926 new_len = min((unsigned long long)io_u->buflen,
1927 (unsigned long long)(zb->wp - io_u->offset));
1928 new_len = new_len / min_bs * min_bs;
1929 if (new_len < io_u->buflen) {
1930 io_u->buflen = new_len;
1931 dprint(FD_IO, "Changed length from %u into %llu\n",
1932 orig_len, io_u->buflen);
1935 assert(zb->start <= io_u->offset);
1936 assert(io_u->offset + io_u->buflen <= zb->wp);
1941 if (io_u->buflen > zbdi->zone_size) {
1942 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
1944 "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n",
1945 f->file_name, io_u->buflen, zbdi->zone_size);
1949 if (!zbd_open_zone(td, f, zone_idx_b)) {
1951 zb = zbd_convert_to_open_zone(td, io_u);
1953 dprint(FD_IO, "%s: can't convert to open zone",
1959 /* Check whether the zone reset threshold has been exceeded */
1960 if (td->o.zrf.u.f) {
1961 if (zbdi->wp_sectors_with_data >= f->io_size * td->o.zrt.u.f &&
1962 zbd_dec_and_reset_write_cnt(td, f))
1966 /* Reset the zone pointer if necessary */
1967 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
1968 assert(td->o.verify == VERIFY_NONE);
1970 * Since previous write requests may have been submitted
1971 * asynchronously and since we will submit the zone
1972 * reset synchronously, wait until previously submitted
1973 * write requests have completed before issuing a
1978 if (zbd_reset_zone(td, f, zb) < 0)
1981 if (zb->capacity < min_bs) {
1982 td_verror(td, EINVAL, "ZCAP is less min_bs");
1983 log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n",
1984 zb->capacity, min_bs);
1989 /* Make writes occur at the write pointer */
1990 assert(!zbd_zone_full(f, zb, min_bs));
1991 io_u->offset = zb->wp;
1992 if (!is_valid_offset(f, io_u->offset)) {
1993 td_verror(td, EINVAL, "invalid WP value");
1994 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
1995 f->file_name, io_u->offset);
2000 * Make sure that the buflen is a multiple of the minimal
2001 * block size. Give up if shrinking would make the request too
2004 new_len = min((unsigned long long)io_u->buflen,
2005 zbd_zone_capacity_end(zb) - io_u->offset);
2006 new_len = new_len / min_bs * min_bs;
2007 if (new_len == io_u->buflen)
2009 if (new_len >= min_bs) {
2010 io_u->buflen = new_len;
2011 dprint(FD_IO, "Changed length from %u into %llu\n",
2012 orig_len, io_u->buflen);
2016 td_verror(td, EIO, "zone remainder too small");
2017 log_err("zone remainder %lld smaller than min block size %"PRIu64"\n",
2018 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
2023 /* Check random trim targets a non-empty zone */
2024 if (!td_random(td) || zb->wp > zb->start)
2027 /* Find out a non-empty zone to trim */
2029 zl = get_zone(f, f->max_zone);
2030 zb = zbd_find_zone(td, io_u, 1, zb, zl);
2032 io_u->offset = zb->start;
2033 dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n",
2034 f->file_name, io_u->offset);
2043 case DDIR_SYNC_FILE_RANGE:
2054 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
2055 assert(!io_u->zbd_queue_io);
2056 assert(!io_u->zbd_put_io);
2058 io_u->zbd_queue_io = zbd_queue_io;
2059 io_u->zbd_put_io = zbd_put_io;
2062 * Since we return with the zone lock still held,
2063 * add an annotation to let Coverity know that it
2066 /* coverity[missing_unlock] */
2071 if (zb && zb->has_wp)
2077 /* Return a string with ZBD statistics */
2078 char *zbd_write_status(const struct thread_stat *ts)
2082 if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0)
2088 * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim
2090 * @td: FIO thread data.
2091 * @io_u: FIO I/O unit.
2093 * It is assumed that z->mutex is already locked.
2094 * Return io_u_completed when reset zone succeeds. Return 0 when the target zone
2095 * does not have write pointer. On error, return negative errno.
2097 int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u)
2099 struct fio_file *f = io_u->file;
2100 struct fio_zone_info *z;
2104 zone_idx = zbd_zone_idx(f, io_u->offset);
2105 z = get_zone(f, zone_idx);
2110 if (io_u->offset != z->start) {
2111 log_err("Trim offset not at zone start (%lld)\n",
2116 ret = zbd_reset_zone((struct thread_data *)td, f, z);
2120 return io_u_completed;