2 * Copyright (C) 2018 Western Digital Corporation or its affiliates.
4 * This file is released under the GPL.
19 #include "oslib/asprintf.h"
25 static bool is_valid_offset(const struct fio_file *f, uint64_t offset)
27 return (uint64_t)(offset - f->file_offset) < f->io_size;
30 static inline unsigned int zbd_zone_idx(const struct fio_file *f,
31 struct fio_zone_info *zone)
33 return zone - f->zbd_info->zone_info;
37 * zbd_offset_to_zone_idx - convert an offset into a zone number
39 * @offset: offset in bytes. If this offset is in the first zone_size bytes
40 * past the disk size then the index of the sentinel is returned.
42 static unsigned int zbd_offset_to_zone_idx(const struct fio_file *f,
47 if (f->zbd_info->zone_size_log2 > 0)
48 zone_idx = offset >> f->zbd_info->zone_size_log2;
50 zone_idx = offset / f->zbd_info->zone_size;
52 return min(zone_idx, f->zbd_info->nr_zones);
56 * zbd_zone_end - Return zone end location
57 * @z: zone info pointer.
59 static inline uint64_t zbd_zone_end(const struct fio_zone_info *z)
65 * zbd_zone_capacity_end - Return zone capacity limit end location
66 * @z: zone info pointer.
68 static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z)
70 return z->start + z->capacity;
74 * zbd_zone_remainder - Return the number of bytes that are still available for
75 * writing before the zone gets full
76 * @z: zone info pointer.
78 static inline uint64_t zbd_zone_remainder(struct fio_zone_info *z)
80 if (z->wp >= zbd_zone_capacity_end(z))
83 return zbd_zone_capacity_end(z) - z->wp;
87 * zbd_zone_full - verify whether a minimum number of bytes remain in a zone
89 * @z: zone info pointer.
90 * @required: minimum number of bytes that must remain in a zone.
92 * The caller must hold z->mutex.
94 static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z,
97 assert((required & 511) == 0);
99 return z->has_wp && required > zbd_zone_remainder(z);
102 static void zone_lock(struct thread_data *td, const struct fio_file *f,
103 struct fio_zone_info *z)
105 struct zoned_block_device_info *zbd = f->zbd_info;
106 uint32_t nz = z - zbd->zone_info;
108 /* A thread should never lock zones outside its working area. */
109 assert(f->min_zone <= nz && nz < f->max_zone);
114 * Lock the io_u target zone. The zone will be unlocked if io_u offset
115 * is changed or when io_u completes and zbd_put_io() executed.
116 * To avoid multiple jobs doing asynchronous I/Os from deadlocking each
117 * other waiting for zone locks when building an io_u batch, first
118 * only trylock the zone. If the zone is already locked by another job,
119 * process the currently queued I/Os so that I/O progress is made and
122 if (pthread_mutex_trylock(&z->mutex) != 0) {
123 if (!td_ioengine_flagged(td, FIO_SYNCIO))
125 pthread_mutex_lock(&z->mutex);
129 static inline void zone_unlock(struct fio_zone_info *z)
134 ret = pthread_mutex_unlock(&z->mutex);
138 static inline struct fio_zone_info *zbd_get_zone(const struct fio_file *f,
139 unsigned int zone_idx)
141 return &f->zbd_info->zone_info[zone_idx];
144 static inline struct fio_zone_info *
145 zbd_offset_to_zone(const struct fio_file *f, uint64_t offset)
147 return zbd_get_zone(f, zbd_offset_to_zone_idx(f, offset));
151 * zbd_get_zoned_model - Get a device zoned model
152 * @td: FIO thread data
153 * @f: FIO file for which to get model information
155 static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f,
156 enum zbd_zoned_model *model)
160 if (f->filetype == FIO_TYPE_PIPE) {
161 log_err("zonemode=zbd does not support pipes\n");
165 /* If regular file, always emulate zones inside the file. */
166 if (f->filetype == FIO_TYPE_FILE) {
171 if (td->io_ops && td->io_ops->get_zoned_model)
172 ret = td->io_ops->get_zoned_model(td, f, model);
174 ret = blkzoned_get_zoned_model(td, f, model);
176 td_verror(td, errno, "get zoned model failed");
177 log_err("%s: get zoned model failed (%d).\n",
178 f->file_name, errno);
185 * zbd_report_zones - Get zone information
186 * @td: FIO thread data.
187 * @f: FIO file for which to get zone information
188 * @offset: offset from which to report zones
189 * @zones: Array of struct zbd_zone
190 * @nr_zones: Size of @zones array
192 * Get zone information into @zones starting from the zone at offset @offset
193 * for the device specified by @f.
195 * Returns the number of zones reported upon success and a negative error code
196 * upon failure. If the zone report is empty, always assume an error (device
197 * problem) and return -EIO.
199 static int zbd_report_zones(struct thread_data *td, struct fio_file *f,
200 uint64_t offset, struct zbd_zone *zones,
201 unsigned int nr_zones)
205 if (td->io_ops && td->io_ops->report_zones)
206 ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones);
208 ret = blkzoned_report_zones(td, f, offset, zones, nr_zones);
210 td_verror(td, errno, "report zones failed");
211 log_err("%s: report zones from sector %"PRIu64" failed (%d).\n",
212 f->file_name, offset >> 9, errno);
213 } else if (ret == 0) {
214 td_verror(td, errno, "Empty zone report");
215 log_err("%s: report zones from sector %"PRIu64" is empty.\n",
216 f->file_name, offset >> 9);
224 * zbd_reset_wp - reset the write pointer of a range of zones
225 * @td: FIO thread data.
226 * @f: FIO file for which to reset zones
227 * @offset: Starting offset of the first zone to reset
228 * @length: Length of the range of zones to reset
230 * Reset the write pointer of all zones in the range @offset...@offset+@length.
231 * Returns 0 upon success and a negative error code upon failure.
233 static int zbd_reset_wp(struct thread_data *td, struct fio_file *f,
234 uint64_t offset, uint64_t length)
238 if (td->io_ops && td->io_ops->reset_wp)
239 ret = td->io_ops->reset_wp(td, f, offset, length);
241 ret = blkzoned_reset_wp(td, f, offset, length);
243 td_verror(td, errno, "resetting wp failed");
244 log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n",
245 f->file_name, length >> 9, offset >> 9, errno);
252 * zbd_reset_zone - reset the write pointer of a single zone
253 * @td: FIO thread data.
254 * @f: FIO file associated with the disk for which to reset a write pointer.
257 * Returns 0 upon success and a negative error code upon failure.
259 * The caller must hold z->mutex.
261 static int zbd_reset_zone(struct thread_data *td, struct fio_file *f,
262 struct fio_zone_info *z)
264 uint64_t offset = z->start;
265 uint64_t length = (z+1)->start - offset;
266 uint64_t data_in_zone = z->wp - z->start;
272 assert(is_valid_offset(f, offset + length - 1));
274 dprint(FD_ZBD, "%s: resetting wp of zone %u.\n",
275 f->file_name, zbd_zone_idx(f, z));
277 switch (f->zbd_info->model) {
279 case ZBD_HOST_MANAGED:
280 ret = zbd_reset_wp(td, f, offset, length);
288 pthread_mutex_lock(&f->zbd_info->mutex);
289 f->zbd_info->sectors_with_data -= data_in_zone;
290 f->zbd_info->wp_sectors_with_data -= data_in_zone;
291 pthread_mutex_unlock(&f->zbd_info->mutex);
296 td->ts.nr_zone_resets++;
302 * zbd_close_zone - Remove a zone from the open zones array.
303 * @td: FIO thread data.
304 * @f: FIO file associated with the disk for which to reset a write pointer.
305 * @zone_idx: Index of the zone to remove.
307 * The caller must hold f->zbd_info->mutex.
309 static void zbd_close_zone(struct thread_data *td, const struct fio_file *f,
310 struct fio_zone_info *z)
317 for (ozi = 0; ozi < f->zbd_info->num_open_zones; ozi++) {
318 if (zbd_get_zone(f, f->zbd_info->open_zones[ozi]) == z)
321 if (ozi == f->zbd_info->num_open_zones)
324 dprint(FD_ZBD, "%s: closing zone %u\n",
325 f->file_name, zbd_zone_idx(f, z));
327 memmove(f->zbd_info->open_zones + ozi,
328 f->zbd_info->open_zones + ozi + 1,
329 (ZBD_MAX_OPEN_ZONES - (ozi + 1)) *
330 sizeof(f->zbd_info->open_zones[0]));
332 f->zbd_info->num_open_zones--;
333 td->num_open_zones--;
338 * zbd_finish_zone - finish the specified zone
339 * @td: FIO thread data.
340 * @f: FIO file for which to finish a zone
341 * @z: Zone to finish.
343 * Finish the zone at @offset with open or close status.
345 static int zbd_finish_zone(struct thread_data *td, struct fio_file *f,
346 struct fio_zone_info *z)
348 uint64_t offset = z->start;
349 uint64_t length = f->zbd_info->zone_size;
352 switch (f->zbd_info->model) {
354 case ZBD_HOST_MANAGED:
355 if (td->io_ops && td->io_ops->finish_zone)
356 ret = td->io_ops->finish_zone(td, f, offset, length);
358 ret = blkzoned_finish_zone(td, f, offset, length);
365 td_verror(td, errno, "finish zone failed");
366 log_err("%s: finish zone at sector %"PRIu64" failed (%d).\n",
367 f->file_name, offset >> 9, errno);
369 z->wp = (z+1)->start;
376 * zbd_reset_zones - Reset a range of zones.
377 * @td: fio thread data.
378 * @f: fio file for which to reset zones
379 * @zb: first zone to reset.
380 * @ze: first zone not to reset.
382 * Returns 0 upon success and 1 upon failure.
384 static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
385 struct fio_zone_info *const zb,
386 struct fio_zone_info *const ze)
388 struct fio_zone_info *z;
389 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
394 dprint(FD_ZBD, "%s: examining zones %u .. %u\n",
395 f->file_name, zbd_zone_idx(f, zb), zbd_zone_idx(f, ze));
397 for (z = zb; z < ze; z++) {
402 pthread_mutex_lock(&f->zbd_info->mutex);
403 zbd_close_zone(td, f, z);
404 pthread_mutex_unlock(&f->zbd_info->mutex);
406 if (z->wp != z->start) {
407 dprint(FD_ZBD, "%s: resetting zone %u\n",
408 f->file_name, zbd_zone_idx(f, z));
409 if (zbd_reset_zone(td, f, z) < 0)
420 * zbd_get_max_open_zones - Get the maximum number of open zones
421 * @td: FIO thread data
422 * @f: FIO file for which to get max open zones
423 * @max_open_zones: Upon success, result will be stored here.
425 * A @max_open_zones value set to zero means no limit.
427 * Returns 0 upon success and a negative error code upon failure.
429 static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f,
430 unsigned int *max_open_zones)
434 if (td->io_ops && td->io_ops->get_max_open_zones)
435 ret = td->io_ops->get_max_open_zones(td, f, max_open_zones);
437 ret = blkzoned_get_max_open_zones(td, f, max_open_zones);
439 td_verror(td, errno, "get max open zones failed");
440 log_err("%s: get max open zones failed (%d).\n",
441 f->file_name, errno);
448 * zbd_open_zone - Add a zone to the array of open zones.
449 * @td: fio thread data.
450 * @f: fio file that has the open zones to add.
451 * @zone_idx: Index of the zone to add.
453 * Open a ZBD zone if it is not already open. Returns true if either the zone
454 * was already open or if the zone was successfully added to the array of open
455 * zones without exceeding the maximum number of open zones. Returns false if
456 * the zone was not already open and opening the zone would cause the zone limit
459 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
460 struct fio_zone_info *z)
462 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
463 struct zoned_block_device_info *zbdi = f->zbd_info;
464 uint32_t zone_idx = zbd_zone_idx(f, z);
467 if (z->cond == ZBD_ZONE_COND_OFFLINE)
471 * Skip full zones with data verification enabled because resetting a
472 * zone causes data loss and hence causes verification to fail.
474 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
478 * zbdi->max_open_zones == 0 means that there is no limit on the maximum
479 * number of open zones. In this case, do no track open zones in
480 * zbdi->open_zones array.
482 if (!zbdi->max_open_zones)
485 pthread_mutex_lock(&zbdi->mutex);
489 * If the zone is going to be completely filled by writes
490 * already in-flight, handle it as a full zone instead of an
493 if (!zbd_zone_remainder(z))
499 /* Zero means no limit */
500 if (td->o.job_max_open_zones > 0 &&
501 td->num_open_zones >= td->o.job_max_open_zones)
503 if (zbdi->num_open_zones >= zbdi->max_open_zones)
506 dprint(FD_ZBD, "%s: opening zone %u\n",
507 f->file_name, zone_idx);
509 zbdi->open_zones[zbdi->num_open_zones++] = zone_idx;
510 td->num_open_zones++;
515 pthread_mutex_unlock(&zbdi->mutex);
519 /* Verify whether direct I/O is used for all host-managed zoned block drives. */
520 static bool zbd_using_direct_io(void)
522 struct thread_data *td;
527 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
529 for_each_file(td, f, j) {
530 if (f->zbd_info && f->filetype == FIO_TYPE_BLOCK &&
531 f->zbd_info->model == ZBD_HOST_MANAGED)
539 /* Whether or not the I/O range for f includes one or more sequential zones */
540 static bool zbd_is_seq_job(struct fio_file *f)
542 uint32_t zone_idx, zone_idx_b, zone_idx_e;
549 zone_idx_b = zbd_offset_to_zone_idx(f, f->file_offset);
551 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size - 1);
552 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
553 if (zbd_get_zone(f, zone_idx)->has_wp)
560 * Verify whether the file offset and size parameters are aligned with zone
561 * boundaries. If the file offset is not aligned, align it down to the start of
562 * the zone containing the start offset and align up the file io_size parameter.
564 static bool zbd_zone_align_file_sizes(struct thread_data *td,
567 const struct fio_zone_info *z;
568 uint64_t new_offset, new_end;
572 if (f->file_offset >= f->real_file_size)
574 if (!zbd_is_seq_job(f))
577 if (!td->o.zone_size) {
578 td->o.zone_size = f->zbd_info->zone_size;
579 if (!td->o.zone_size) {
580 log_err("%s: invalid 0 zone size\n",
584 } else if (td->o.zone_size != f->zbd_info->zone_size) {
585 log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n",
586 f->file_name, td->o.zone_size,
587 f->zbd_info->zone_size);
591 if (td->o.zone_skip % td->o.zone_size) {
592 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
593 f->file_name, td->o.zone_skip,
598 z = zbd_offset_to_zone(f, f->file_offset);
599 if ((f->file_offset != z->start) &&
600 (td->o.td_ddir != TD_DDIR_READ)) {
601 new_offset = zbd_zone_end(z);
602 if (new_offset >= f->file_offset + f->io_size) {
603 log_info("%s: io_size must be at least one zone\n",
607 log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n",
608 f->file_name, f->file_offset,
610 f->io_size -= (new_offset - f->file_offset);
611 f->file_offset = new_offset;
614 z = zbd_offset_to_zone(f, f->file_offset + f->io_size);
616 if ((td->o.td_ddir != TD_DDIR_READ) &&
617 (f->file_offset + f->io_size != new_end)) {
618 if (new_end <= f->file_offset) {
619 log_info("%s: io_size must be at least one zone\n",
623 log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n",
624 f->file_name, f->io_size,
625 new_end - f->file_offset);
626 f->io_size = new_end - f->file_offset;
633 * Verify whether offset and size parameters are aligned with zone boundaries.
635 static bool zbd_verify_sizes(void)
637 struct thread_data *td;
642 for_each_file(td, f, j) {
643 if (!zbd_zone_align_file_sizes(td, f))
651 static bool zbd_verify_bs(void)
653 struct thread_data *td;
659 (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] ||
660 td->o.bssplit_nr[DDIR_TRIM])) {
661 log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n");
664 for_each_file(td, f, j) {
670 zone_size = f->zbd_info->zone_size;
671 if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) {
672 log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n",
673 f->file_name, td->o.bs[DDIR_TRIM],
682 static int ilog2(uint64_t i)
694 * Initialize f->zbd_info for devices that are not zoned block devices. This
695 * allows to execute a ZBD workload against a non-ZBD device.
697 static int init_zone_info(struct thread_data *td, struct fio_file *f)
700 struct fio_zone_info *p;
701 uint64_t zone_size = td->o.zone_size;
702 uint64_t zone_capacity = td->o.zone_capacity;
703 struct zoned_block_device_info *zbd_info = NULL;
706 if (zone_size == 0) {
707 log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n",
712 if (zone_size < 512) {
713 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
718 if (zone_capacity == 0)
719 zone_capacity = zone_size;
721 if (zone_capacity > zone_size) {
722 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
723 f->file_name, td->o.zone_capacity, td->o.zone_size);
727 if (f->real_file_size < zone_size) {
728 log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n",
729 f->file_name, f->real_file_size, zone_size);
733 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
734 zbd_info = scalloc(1, sizeof(*zbd_info) +
735 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
739 mutex_init_pshared(&zbd_info->mutex);
740 zbd_info->refcount = 1;
741 p = &zbd_info->zone_info[0];
742 for (i = 0; i < nr_zones; i++, p++) {
743 mutex_init_pshared_with_type(&p->mutex,
744 PTHREAD_MUTEX_RECURSIVE);
745 p->start = i * zone_size;
747 p->type = ZBD_ZONE_TYPE_SWR;
748 p->cond = ZBD_ZONE_COND_EMPTY;
749 p->capacity = zone_capacity;
753 p->start = nr_zones * zone_size;
755 f->zbd_info = zbd_info;
756 f->zbd_info->zone_size = zone_size;
757 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
758 ilog2(zone_size) : 0;
759 f->zbd_info->nr_zones = nr_zones;
764 * Maximum number of zones to report in one operation.
766 #define ZBD_REPORT_MAX_ZONES 8192U
769 * Parse the device zone report and store it in f->zbd_info. Must be called
770 * only for devices that are zoned, namely those with a model != ZBD_NONE.
772 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
775 struct zbd_zone *zones, *z;
776 struct fio_zone_info *p;
777 uint64_t zone_size, offset;
778 struct zoned_block_device_info *zbd_info = NULL;
779 int i, j, ret = -ENOMEM;
781 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
785 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
788 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
793 zone_size = zones[0].len;
794 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
796 if (td->o.zone_size == 0) {
797 td->o.zone_size = zone_size;
798 } else if (td->o.zone_size != zone_size) {
799 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
800 f->file_name, td->o.zone_size, zone_size);
805 dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n",
806 f->file_name, nr_zones, zone_size / 1024);
808 zbd_info = scalloc(1, sizeof(*zbd_info) +
809 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
812 mutex_init_pshared(&zbd_info->mutex);
813 zbd_info->refcount = 1;
814 p = &zbd_info->zone_info[0];
815 for (offset = 0, j = 0; j < nr_zones;) {
817 for (i = 0; i < nrz; i++, j++, z++, p++) {
818 mutex_init_pshared_with_type(&p->mutex,
819 PTHREAD_MUTEX_RECURSIVE);
821 p->capacity = z->capacity;
824 case ZBD_ZONE_COND_NOT_WP:
825 case ZBD_ZONE_COND_FULL:
826 p->wp = p->start + p->capacity;
829 assert(z->start <= z->wp);
830 assert(z->wp <= z->start + zone_size);
836 case ZBD_ZONE_TYPE_SWR:
845 if (j > 0 && p->start != p[-1].start + zone_size) {
846 log_info("%s: invalid zone data\n",
853 offset = z->start + z->len;
857 nrz = zbd_report_zones(td, f, offset, zones,
858 min((uint32_t)(nr_zones - j),
859 ZBD_REPORT_MAX_ZONES));
862 log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n",
863 offset, f->file_name, -ret);
869 zbd_info->zone_info[nr_zones].start = offset;
871 f->zbd_info = zbd_info;
872 f->zbd_info->zone_size = zone_size;
873 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
874 ilog2(zone_size) : 0;
875 f->zbd_info->nr_zones = nr_zones;
885 static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f)
887 struct zoned_block_device_info *zbd = f->zbd_info;
888 unsigned int max_open_zones;
891 if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) {
892 /* Only host-managed devices have a max open limit */
893 zbd->max_open_zones = td->o.max_open_zones;
897 /* If host-managed, get the max open limit */
898 ret = zbd_get_max_open_zones(td, f, &max_open_zones);
902 if (!max_open_zones) {
903 /* No device limit */
904 zbd->max_open_zones = td->o.max_open_zones;
905 } else if (!td->o.max_open_zones) {
906 /* No user limit. Set limit to device limit */
907 zbd->max_open_zones = max_open_zones;
908 } else if (td->o.max_open_zones <= max_open_zones) {
909 /* Both user limit and dev limit. User limit not too large */
910 zbd->max_open_zones = td->o.max_open_zones;
912 /* Both user limit and dev limit. User limit too large */
913 td_verror(td, EINVAL,
914 "Specified --max_open_zones is too large");
915 log_err("Specified --max_open_zones (%d) is larger than max (%u)\n",
916 td->o.max_open_zones, max_open_zones);
921 /* Ensure that the limit is not larger than FIO's internal limit */
922 if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) {
923 td_verror(td, EINVAL, "'max_open_zones' value is too large");
924 log_err("'max_open_zones' value is larger than %u\n",
929 dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n",
930 f->file_name, zbd->max_open_zones);
936 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
938 * Returns 0 upon success and a negative error code upon failure.
940 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
942 enum zbd_zoned_model zbd_model;
945 assert(td->o.zone_mode == ZONE_MODE_ZBD);
947 ret = zbd_get_zoned_model(td, f, &zbd_model);
953 case ZBD_HOST_MANAGED:
954 ret = parse_zone_info(td, f);
959 ret = init_zone_info(td, f);
964 td_verror(td, EINVAL, "Unsupported zoned model");
965 log_err("Unsupported zoned model\n");
970 f->zbd_info->model = zbd_model;
972 ret = zbd_set_max_open_zones(td, f);
974 zbd_free_zone_info(f);
981 void zbd_free_zone_info(struct fio_file *f)
987 pthread_mutex_lock(&f->zbd_info->mutex);
988 refcount = --f->zbd_info->refcount;
989 pthread_mutex_unlock(&f->zbd_info->mutex);
991 assert((int32_t)refcount >= 0);
998 * Initialize f->zbd_info.
1000 * Returns 0 upon success and a negative error code upon failure.
1002 * Note: this function can only work correctly if it is called before the first
1005 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
1007 struct thread_data *td2;
1008 struct fio_file *f2;
1011 for_each_td(td2, i) {
1012 for_each_file(td2, f2, j) {
1013 if (td2 == td && f2 == file)
1015 if (!f2->zbd_info ||
1016 strcmp(f2->file_name, file->file_name) != 0)
1018 file->zbd_info = f2->zbd_info;
1019 file->zbd_info->refcount++;
1024 ret = zbd_create_zone_info(td, file);
1026 td_verror(td, -ret, "zbd_create_zone_info() failed");
1031 int zbd_init_files(struct thread_data *td)
1036 for_each_file(td, f, i) {
1037 if (zbd_init_zone_info(td, f))
1044 void zbd_recalc_options_with_zone_granularity(struct thread_data *td)
1049 for_each_file(td, f, i) {
1050 struct zoned_block_device_info *zbd = f->zbd_info;
1053 /* zonemode=strided doesn't get per-file zone size. */
1054 zone_size = zbd ? zbd->zone_size : td->o.zone_size;
1058 if (td->o.size_nz > 0)
1059 td->o.size = td->o.size_nz * zone_size;
1060 if (td->o.io_size_nz > 0)
1061 td->o.io_size = td->o.io_size_nz * zone_size;
1062 if (td->o.start_offset_nz > 0)
1063 td->o.start_offset = td->o.start_offset_nz * zone_size;
1064 if (td->o.offset_increment_nz > 0)
1065 td->o.offset_increment =
1066 td->o.offset_increment_nz * zone_size;
1067 if (td->o.zone_skip_nz > 0)
1068 td->o.zone_skip = td->o.zone_skip_nz * zone_size;
1072 int zbd_setup_files(struct thread_data *td)
1077 if (!zbd_using_direct_io()) {
1078 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
1082 if (!zbd_verify_sizes())
1085 if (!zbd_verify_bs())
1088 for_each_file(td, f, i) {
1089 struct zoned_block_device_info *zbd = f->zbd_info;
1090 struct fio_zone_info *z;
1095 f->min_zone = zbd_offset_to_zone_idx(f, f->file_offset);
1097 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size);
1100 * When all zones in the I/O range are conventional, io_size
1101 * can be smaller than zone size, making min_zone the same
1102 * as max_zone. This is why the assert below needs to be made
1105 if (zbd_is_seq_job(f))
1106 assert(f->min_zone < f->max_zone);
1108 if (td->o.max_open_zones > 0 &&
1109 zbd->max_open_zones != td->o.max_open_zones) {
1110 log_err("Different 'max_open_zones' values\n");
1115 * The per job max open zones limit cannot be used without a
1116 * global max open zones limit. (As the tracking of open zones
1117 * is disabled when there is no global max open zones limit.)
1119 if (td->o.job_max_open_zones && !zbd->max_open_zones) {
1120 log_err("'job_max_open_zones' cannot be used without a global open zones limit\n");
1125 * zbd->max_open_zones is the global limit shared for all jobs
1126 * that target the same zoned block device. Force sync the per
1127 * thread global limit with the actual global limit. (The real
1128 * per thread/job limit is stored in td->o.job_max_open_zones).
1130 td->o.max_open_zones = zbd->max_open_zones;
1132 for (zi = f->min_zone; zi < f->max_zone; zi++) {
1133 z = &zbd->zone_info[zi];
1134 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
1135 z->cond != ZBD_ZONE_COND_EXP_OPEN)
1137 if (zbd_open_zone(td, f, z))
1140 * If the number of open zones exceeds specified limits,
1141 * reset all extra open zones.
1143 if (zbd_reset_zone(td, f, z) < 0) {
1144 log_err("Failed to reest zone %d\n", zi);
1154 * Reset zbd_info.write_cnt, the counter that counts down towards the next
1157 static void _zbd_reset_write_cnt(const struct thread_data *td,
1158 const struct fio_file *f)
1160 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
1162 f->zbd_info->write_cnt = td->o.zrf.u.f ?
1163 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
1166 static void zbd_reset_write_cnt(const struct thread_data *td,
1167 const struct fio_file *f)
1169 pthread_mutex_lock(&f->zbd_info->mutex);
1170 _zbd_reset_write_cnt(td, f);
1171 pthread_mutex_unlock(&f->zbd_info->mutex);
1174 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
1175 const struct fio_file *f)
1177 uint32_t write_cnt = 0;
1179 pthread_mutex_lock(&f->zbd_info->mutex);
1180 assert(f->zbd_info->write_cnt);
1181 if (f->zbd_info->write_cnt)
1182 write_cnt = --f->zbd_info->write_cnt;
1184 _zbd_reset_write_cnt(td, f);
1185 pthread_mutex_unlock(&f->zbd_info->mutex);
1187 return write_cnt == 0;
1195 /* Calculate the number of sectors with data (swd) and perform action 'a' */
1196 static uint64_t zbd_process_swd(struct thread_data *td,
1197 const struct fio_file *f, enum swd_action a)
1199 struct fio_zone_info *zb, *ze, *z;
1201 uint64_t wp_swd = 0;
1203 zb = zbd_get_zone(f, f->min_zone);
1204 ze = zbd_get_zone(f, f->max_zone);
1205 for (z = zb; z < ze; z++) {
1207 zone_lock(td, f, z);
1208 wp_swd += z->wp - z->start;
1210 swd += z->wp - z->start;
1213 pthread_mutex_lock(&f->zbd_info->mutex);
1216 assert(f->zbd_info->sectors_with_data == swd);
1217 assert(f->zbd_info->wp_sectors_with_data == wp_swd);
1220 f->zbd_info->sectors_with_data = swd;
1221 f->zbd_info->wp_sectors_with_data = wp_swd;
1224 pthread_mutex_unlock(&f->zbd_info->mutex);
1226 for (z = zb; z < ze; z++)
1234 * The swd check is useful for debugging but takes too much time to leave
1235 * it enabled all the time. Hence it is disabled by default.
1237 static const bool enable_check_swd = false;
1239 /* Check whether the values of zbd_info.*sectors_with_data are correct. */
1240 static void zbd_check_swd(struct thread_data *td, const struct fio_file *f)
1242 if (!enable_check_swd)
1245 zbd_process_swd(td, f, CHECK_SWD);
1248 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
1250 struct fio_zone_info *zb, *ze;
1252 bool verify_data_left = false;
1254 if (!f->zbd_info || !td_write(td))
1257 zb = zbd_get_zone(f, f->min_zone);
1258 ze = zbd_get_zone(f, f->max_zone);
1259 swd = zbd_process_swd(td, f, SET_SWD);
1261 dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n",
1262 __func__, f->file_name, swd);
1265 * If data verification is enabled reset the affected zones before
1266 * writing any data to avoid that a zone reset has to be issued while
1267 * writing data, which causes data loss.
1269 if (td->o.verify != VERIFY_NONE) {
1270 verify_data_left = td->runstate == TD_VERIFYING ||
1271 td->io_hist_len || td->verify_batch;
1272 if (td->io_hist_len && td->o.verify_backlog)
1274 td->io_hist_len % td->o.verify_backlog;
1275 if (!verify_data_left)
1276 zbd_reset_zones(td, f, zb, ze);
1279 zbd_reset_write_cnt(td, f);
1282 /* Return random zone index for one of the open zones. */
1283 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1284 const struct io_u *io_u)
1286 return (io_u->offset - f->file_offset) *
1287 f->zbd_info->num_open_zones / f->io_size;
1290 static bool any_io_in_flight(void)
1292 struct thread_data *td;
1295 for_each_td(td, i) {
1296 if (td->io_u_in_flight)
1304 * Modify the offset of an I/O unit that does not refer to an open zone such
1305 * that it refers to an open zone. Close an open zone and open a new zone if
1306 * necessary. The open zone is searched across sequential zones.
1307 * This algorithm can only work correctly if all write pointers are
1308 * a multiple of the fio block size. The caller must neither hold z->mutex
1309 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1311 static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
1314 const uint64_t min_bs = td->o.min_bs[io_u->ddir];
1315 struct fio_file *f = io_u->file;
1316 struct zoned_block_device_info *zbdi = f->zbd_info;
1317 struct fio_zone_info *z;
1318 unsigned int open_zone_idx = -1;
1319 uint32_t zone_idx, new_zone_idx;
1321 bool wait_zone_close;
1323 bool should_retry = true;
1325 assert(is_valid_offset(f, io_u->offset));
1327 if (zbdi->max_open_zones || td->o.job_max_open_zones) {
1329 * This statement accesses zbdi->open_zones[] on purpose
1332 zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)];
1334 zone_idx = zbd_offset_to_zone_idx(f, io_u->offset);
1336 if (zone_idx < f->min_zone)
1337 zone_idx = f->min_zone;
1338 else if (zone_idx >= f->max_zone)
1339 zone_idx = f->max_zone - 1;
1342 "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1343 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1346 * Since z->mutex is the outer lock and zbdi->mutex the inner
1347 * lock it can happen that the state of the zone with index zone_idx
1348 * has changed after 'z' has been assigned and before zbdi->mutex
1349 * has been obtained. Hence the loop.
1354 z = zbd_get_zone(f, zone_idx);
1356 zone_lock(td, f, z);
1358 pthread_mutex_lock(&zbdi->mutex);
1361 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1362 zbdi->max_open_zones == 0 &&
1363 td->o.job_max_open_zones == 0)
1365 if (zbdi->num_open_zones == 0) {
1366 dprint(FD_ZBD, "%s(%s): no zones are open\n",
1367 __func__, f->file_name);
1368 goto open_other_zone;
1373 * List of opened zones is per-device, shared across all
1374 * threads. Start with quasi-random candidate zone. Ignore
1375 * zones which don't belong to thread's offset/size area.
1377 open_zone_idx = pick_random_zone_idx(f, io_u);
1378 assert(!open_zone_idx ||
1379 open_zone_idx < zbdi->num_open_zones);
1380 tmp_idx = open_zone_idx;
1382 for (i = 0; i < zbdi->num_open_zones; i++) {
1385 if (tmp_idx >= zbdi->num_open_zones)
1387 tmpz = zbdi->open_zones[tmp_idx];
1388 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1389 open_zone_idx = tmp_idx;
1390 goto found_candidate_zone;
1396 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1397 __func__, f->file_name);
1399 pthread_mutex_unlock(&zbdi->mutex);
1406 found_candidate_zone:
1407 new_zone_idx = zbdi->open_zones[open_zone_idx];
1408 if (new_zone_idx == zone_idx)
1410 zone_idx = new_zone_idx;
1412 pthread_mutex_unlock(&zbdi->mutex);
1418 /* Both z->mutex and zbdi->mutex are held. */
1421 if (zbd_zone_remainder(z) >= min_bs) {
1422 pthread_mutex_unlock(&zbdi->mutex);
1427 /* Check if number of open zones reaches one of limits. */
1429 zbdi->num_open_zones == f->max_zone - f->min_zone ||
1430 (zbdi->max_open_zones &&
1431 zbdi->num_open_zones == zbdi->max_open_zones) ||
1432 (td->o.job_max_open_zones &&
1433 td->num_open_zones == td->o.job_max_open_zones);
1435 pthread_mutex_unlock(&zbdi->mutex);
1437 /* Only z->mutex is held. */
1440 * When number of open zones reaches to one of limits, wait for
1441 * zone close before opening a new zone.
1443 if (wait_zone_close) {
1445 "%s(%s): quiesce to allow open zones to close\n",
1446 __func__, f->file_name);
1451 /* Zone 'z' is full, so try to open a new zone. */
1452 for (i = f->io_size / zbdi->zone_size; i > 0; i--) {
1457 if (!is_valid_offset(f, z->start)) {
1459 zone_idx = f->min_zone;
1460 z = zbd_get_zone(f, zone_idx);
1462 assert(is_valid_offset(f, z->start));
1465 zone_lock(td, f, z);
1468 if (zbd_open_zone(td, f, z))
1472 /* Only z->mutex is held. */
1474 /* Check whether the write fits in any of the already opened zones. */
1475 pthread_mutex_lock(&zbdi->mutex);
1476 for (i = 0; i < zbdi->num_open_zones; i++) {
1477 zone_idx = zbdi->open_zones[i];
1478 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1480 pthread_mutex_unlock(&zbdi->mutex);
1483 z = zbd_get_zone(f, zone_idx);
1485 zone_lock(td, f, z);
1486 if (zbd_zone_remainder(z) >= min_bs)
1488 pthread_mutex_lock(&zbdi->mutex);
1492 * When any I/O is in-flight or when all I/Os in-flight get completed,
1493 * the I/Os might have closed zones then retry the steps to open a zone.
1494 * Before retry, call io_u_quiesce() to complete in-flight writes.
1496 in_flight = any_io_in_flight();
1497 if (in_flight || should_retry) {
1499 "%s(%s): wait zone close and retry open zones\n",
1500 __func__, f->file_name);
1501 pthread_mutex_unlock(&zbdi->mutex);
1504 zone_lock(td, f, z);
1505 should_retry = in_flight;
1509 pthread_mutex_unlock(&zbdi->mutex);
1513 dprint(FD_ZBD, "%s(%s): did not open another zone\n",
1514 __func__, f->file_name);
1519 dprint(FD_ZBD, "%s(%s): returning zone %d\n",
1520 __func__, f->file_name, zone_idx);
1522 io_u->offset = z->start;
1524 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1529 /* The caller must hold z->mutex. */
1530 static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td,
1532 struct fio_zone_info *z)
1534 const struct fio_file *f = io_u->file;
1535 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
1537 if (!zbd_open_zone(td, f, z)) {
1539 z = zbd_convert_to_open_zone(td, io_u);
1543 if (z->verify_block * min_bs >= z->capacity) {
1544 log_err("%s: %d * %"PRIu64" >= %"PRIu64"\n",
1545 f->file_name, z->verify_block, min_bs, z->capacity);
1547 * If the assertion below fails during a test run, adding
1548 * "--experimental_verify=1" to the command line may help.
1553 io_u->offset = z->start + z->verify_block * min_bs;
1554 if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1555 log_err("%s: %llu + %llu >= %"PRIu64"\n",
1556 f->file_name, io_u->offset, io_u->buflen,
1557 zbd_zone_capacity_end(z));
1560 z->verify_block += io_u->buflen / min_bs;
1566 * Find another zone which has @min_bytes of readable data. Search in zones
1567 * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf.
1569 * Either returns NULL or returns a zone pointer. When the zone has write
1570 * pointer, hold the mutex for the zone.
1572 static struct fio_zone_info *
1573 zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes,
1574 struct fio_zone_info *zb, struct fio_zone_info *zl)
1576 struct fio_file *f = io_u->file;
1577 struct fio_zone_info *z1, *z2;
1578 const struct fio_zone_info *const zf = zbd_get_zone(f, f->min_zone);
1581 * Skip to the next non-empty zone in case of sequential I/O and to
1582 * the nearest non-empty zone in case of random I/O.
1584 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1585 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1587 zone_lock(td, f, z1);
1588 if (z1->start + min_bytes <= z1->wp)
1592 } else if (!td_random(td)) {
1596 if (td_random(td) && z2 >= zf &&
1597 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1599 zone_lock(td, f, z2);
1600 if (z2->start + min_bytes <= z2->wp)
1608 "%s: no zone has %"PRIu64" bytes of readable data\n",
1609 f->file_name, min_bytes);
1615 * zbd_end_zone_io - update zone status at command completion
1617 * @z: zone info pointer
1619 * If the write command made the zone full, close it.
1621 * The caller must hold z->mutex.
1623 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1624 struct fio_zone_info *z)
1626 const struct fio_file *f = io_u->file;
1628 if (io_u->ddir == DDIR_WRITE &&
1629 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1630 pthread_mutex_lock(&f->zbd_info->mutex);
1631 zbd_close_zone(td, f, z);
1632 pthread_mutex_unlock(&f->zbd_info->mutex);
1637 * zbd_queue_io - update the write pointer of a sequential zone
1639 * @success: Whether or not the I/O unit has been queued successfully
1640 * @q: queueing status (busy, completed or queued).
1642 * For write and trim operations, update the write pointer of the I/O unit
1645 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1648 const struct fio_file *f = io_u->file;
1649 struct zoned_block_device_info *zbd_info = f->zbd_info;
1650 struct fio_zone_info *z;
1655 z = zbd_offset_to_zone(f, io_u->offset);
1662 "%s: queued I/O (%lld, %llu) for zone %u\n",
1663 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1665 switch (io_u->ddir) {
1667 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1668 zbd_zone_capacity_end(z));
1671 * z->wp > zone_end means that one or more I/O errors
1674 pthread_mutex_lock(&zbd_info->mutex);
1675 if (z->wp <= zone_end) {
1676 zbd_info->sectors_with_data += zone_end - z->wp;
1677 zbd_info->wp_sectors_with_data += zone_end - z->wp;
1679 pthread_mutex_unlock(&zbd_info->mutex);
1686 if (q == FIO_Q_COMPLETED && !io_u->error)
1687 zbd_end_zone_io(td, io_u, z);
1690 if (!success || q != FIO_Q_QUEUED) {
1691 /* BUSY or COMPLETED: unlock the zone */
1693 io_u->zbd_put_io = NULL;
1698 * zbd_put_io - Unlock an I/O unit target zone lock
1701 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1703 const struct fio_file *f = io_u->file;
1704 struct zoned_block_device_info *zbd_info = f->zbd_info;
1705 struct fio_zone_info *z;
1709 z = zbd_offset_to_zone(f, io_u->offset);
1713 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1714 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1716 zbd_end_zone_io(td, io_u, z);
1719 zbd_check_swd(td, f);
1723 * Windows and MacOS do not define this.
1726 #define EREMOTEIO 121 /* POSIX value */
1729 bool zbd_unaligned_write(int error_code)
1731 switch (error_code) {
1740 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1741 * @td: FIO thread data.
1742 * @io_u: FIO I/O unit.
1744 * For sequential workloads, change the file offset to skip zoneskip bytes when
1745 * no more IO can be performed in the current zone.
1746 * - For read workloads, zoneskip is applied when the io has reached the end of
1747 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1748 * - For write workloads, zoneskip is applied when the zone is full.
1749 * This applies only to read and write operations.
1751 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1753 struct fio_file *f = io_u->file;
1754 enum fio_ddir ddir = io_u->ddir;
1755 struct fio_zone_info *z;
1757 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1758 assert(td->o.zone_size);
1759 assert(f->zbd_info);
1761 z = zbd_offset_to_zone(f, f->last_pos[ddir]);
1764 * When the zone capacity is smaller than the zone size and the I/O is
1765 * sequential write, skip to zone end if the latest position is at the
1766 * zone capacity limit.
1768 if (z->capacity < f->zbd_info->zone_size &&
1769 !td_random(td) && ddir == DDIR_WRITE &&
1770 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1772 "%s: Jump from zone capacity limit to zone end:"
1773 " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n",
1774 f->file_name, f->last_pos[ddir],
1775 zbd_zone_end(z), zbd_zone_idx(f, z), z->capacity);
1776 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1777 f->last_pos[ddir] = zbd_zone_end(z);
1781 * zone_skip is valid only for sequential workloads.
1783 if (td_random(td) || !td->o.zone_skip)
1787 * It is time to switch to a new zone if:
1788 * - zone_bytes == zone_size bytes have already been accessed
1789 * - The last position reached the end of the current zone.
1790 * - For reads with td->o.read_beyond_wp == false, the last position
1791 * reached the zone write pointer.
1793 if (td->zone_bytes >= td->o.zone_size ||
1794 f->last_pos[ddir] >= zbd_zone_end(z) ||
1795 (ddir == DDIR_READ &&
1796 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1801 f->file_offset += td->o.zone_size + td->o.zone_skip;
1804 * Wrap from the beginning, if we exceed the file size
1806 if (f->file_offset >= f->real_file_size)
1807 f->file_offset = get_start_offset(td, f);
1809 f->last_pos[ddir] = f->file_offset;
1810 td->io_skip_bytes += td->o.zone_skip;
1815 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1817 * @td: FIO thread data.
1818 * @io_u: FIO I/O unit.
1819 * @ddir: I/O direction before adjustment.
1821 * Return adjusted I/O direction.
1823 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1827 * In case read direction is chosen for the first random I/O, fio with
1828 * zonemode=zbd stops because no data can be read from zoned block
1829 * devices with all empty zones. Overwrite the first I/O direction as
1830 * write to make sure data to read exists.
1832 assert(io_u->file->zbd_info);
1833 if (ddir != DDIR_READ || !td_rw(td))
1836 if (io_u->file->zbd_info->sectors_with_data ||
1837 td->o.read_beyond_wp)
1844 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1845 * @td: FIO thread data.
1846 * @io_u: FIO I/O unit.
1848 * Locking strategy: returns with z->mutex locked if and only if z refers
1849 * to a sequential zone and if io_u_accept is returned. z is the zone that
1850 * corresponds to io_u->offset at the end of this function.
1852 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1854 struct fio_file *f = io_u->file;
1855 struct zoned_block_device_info *zbdi = f->zbd_info;
1856 struct fio_zone_info *zb, *zl, *orig_zb;
1857 uint32_t orig_len = io_u->buflen;
1858 uint64_t min_bs = td->o.min_bs[io_u->ddir];
1864 assert(is_valid_offset(f, io_u->offset));
1865 assert(io_u->buflen);
1867 zb = zbd_offset_to_zone(f, io_u->offset);
1871 /* Accept non-write I/Os for conventional zones. */
1872 if (io_u->ddir != DDIR_WRITE)
1876 * Make sure that writes to conventional zones
1877 * don't cross over to any sequential zones.
1879 if (!(zb + 1)->has_wp ||
1880 io_u->offset + io_u->buflen <= (zb + 1)->start)
1883 if (io_u->offset + min_bs > (zb + 1)->start) {
1885 "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n",
1886 f->file_name, io_u->offset,
1887 min_bs, (zb + 1)->start);
1889 zb->start + (zb + 1)->start - io_u->offset;
1890 new_len = min(io_u->buflen,
1891 (zb + 1)->start - io_u->offset);
1893 new_len = (zb + 1)->start - io_u->offset;
1896 io_u->buflen = new_len / min_bs * min_bs;
1902 * Accept the I/O offset for reads if reading beyond the write pointer
1905 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1906 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1909 zbd_check_swd(td, f);
1911 zone_lock(td, f, zb);
1913 switch (io_u->ddir) {
1915 if (td->runstate == TD_VERIFYING && td_write(td)) {
1916 zb = zbd_replay_write_order(td, io_u, zb);
1921 * Check that there is enough written data in the zone to do an
1922 * I/O of at least min_bs B. If there isn't, find a new zone for
1925 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1926 zb->wp - zb->start : 0;
1927 if (range < min_bs ||
1928 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1930 zl = zbd_get_zone(f, f->max_zone);
1931 zb = zbd_find_zone(td, io_u, min_bs, zb, zl);
1934 "%s: zbd_find_zone(%lld, %llu) failed\n",
1935 f->file_name, io_u->offset,
1940 * zbd_find_zone() returned a zone with a range of at
1943 range = zb->wp - zb->start;
1944 assert(range >= min_bs);
1947 io_u->offset = zb->start;
1951 * Make sure the I/O is within the zone valid data range while
1952 * maximizing the I/O size and preserving randomness.
1954 if (range <= io_u->buflen)
1955 io_u->offset = zb->start;
1956 else if (td_random(td))
1957 io_u->offset = zb->start +
1958 ((io_u->offset - orig_zb->start) %
1959 (range - io_u->buflen)) / min_bs * min_bs;
1962 * When zbd_find_zone() returns a conventional zone,
1963 * we can simply accept the new i/o offset here.
1969 * Make sure the I/O does not cross over the zone wp position.
1971 new_len = min((unsigned long long)io_u->buflen,
1972 (unsigned long long)(zb->wp - io_u->offset));
1973 new_len = new_len / min_bs * min_bs;
1974 if (new_len < io_u->buflen) {
1975 io_u->buflen = new_len;
1976 dprint(FD_IO, "Changed length from %u into %llu\n",
1977 orig_len, io_u->buflen);
1980 assert(zb->start <= io_u->offset);
1981 assert(io_u->offset + io_u->buflen <= zb->wp);
1986 if (io_u->buflen > zbdi->zone_size) {
1987 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
1989 "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n",
1990 f->file_name, io_u->buflen, zbdi->zone_size);
1995 if (zbd_zone_remainder(zb) > 0 &&
1996 zbd_zone_remainder(zb) < min_bs) {
1997 pthread_mutex_lock(&f->zbd_info->mutex);
1998 zbd_close_zone(td, f, zb);
1999 pthread_mutex_unlock(&f->zbd_info->mutex);
2001 "%s: finish zone %d\n",
2002 f->file_name, zbd_zone_idx(f, zb));
2004 zbd_finish_zone(td, f, zb);
2005 if (zbd_zone_idx(f, zb) + 1 >= f->max_zone) {
2011 /* Find the next write pointer zone */
2014 if (zbd_zone_idx(f, zb) >= f->max_zone)
2015 zb = zbd_get_zone(f, f->min_zone);
2016 } while (!zb->has_wp);
2018 zone_lock(td, f, zb);
2021 if (!zbd_open_zone(td, f, zb)) {
2023 zb = zbd_convert_to_open_zone(td, io_u);
2025 dprint(FD_IO, "%s: can't convert to open zone",
2031 if (zbd_zone_remainder(zb) > 0 &&
2032 zbd_zone_remainder(zb) < min_bs)
2035 /* Check whether the zone reset threshold has been exceeded */
2036 if (td->o.zrf.u.f) {
2037 if (zbdi->wp_sectors_with_data >= f->io_size * td->o.zrt.u.f &&
2038 zbd_dec_and_reset_write_cnt(td, f))
2042 /* Reset the zone pointer if necessary */
2043 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
2044 if (td->o.verify != VERIFY_NONE) {
2046 * Unset io-u->file to tell get_next_verify()
2047 * that this IO is not requeue.
2050 if (!get_next_verify(td, io_u)) {
2058 * Since previous write requests may have been submitted
2059 * asynchronously and since we will submit the zone
2060 * reset synchronously, wait until previously submitted
2061 * write requests have completed before issuing a
2066 if (zbd_reset_zone(td, f, zb) < 0)
2069 if (zb->capacity < min_bs) {
2070 td_verror(td, EINVAL, "ZCAP is less min_bs");
2071 log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n",
2072 zb->capacity, min_bs);
2077 /* Make writes occur at the write pointer */
2078 assert(!zbd_zone_full(f, zb, min_bs));
2079 io_u->offset = zb->wp;
2080 if (!is_valid_offset(f, io_u->offset)) {
2081 td_verror(td, EINVAL, "invalid WP value");
2082 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
2083 f->file_name, io_u->offset);
2088 * Make sure that the buflen is a multiple of the minimal
2089 * block size. Give up if shrinking would make the request too
2092 new_len = min((unsigned long long)io_u->buflen,
2093 zbd_zone_capacity_end(zb) - io_u->offset);
2094 new_len = new_len / min_bs * min_bs;
2095 if (new_len == io_u->buflen)
2097 if (new_len >= min_bs) {
2098 io_u->buflen = new_len;
2099 dprint(FD_IO, "Changed length from %u into %llu\n",
2100 orig_len, io_u->buflen);
2104 td_verror(td, EIO, "zone remainder too small");
2105 log_err("zone remainder %lld smaller than min block size %"PRIu64"\n",
2106 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
2111 /* Check random trim targets a non-empty zone */
2112 if (!td_random(td) || zb->wp > zb->start)
2115 /* Find out a non-empty zone to trim */
2117 zl = zbd_get_zone(f, f->max_zone);
2118 zb = zbd_find_zone(td, io_u, 1, zb, zl);
2120 io_u->offset = zb->start;
2121 dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n",
2122 f->file_name, io_u->offset);
2131 case DDIR_SYNC_FILE_RANGE:
2142 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
2143 assert(!io_u->zbd_queue_io);
2144 assert(!io_u->zbd_put_io);
2146 io_u->zbd_queue_io = zbd_queue_io;
2147 io_u->zbd_put_io = zbd_put_io;
2150 * Since we return with the zone lock still held,
2151 * add an annotation to let Coverity know that it
2154 /* coverity[missing_unlock] */
2159 if (zb && zb->has_wp)
2165 /* Return a string with ZBD statistics */
2166 char *zbd_write_status(const struct thread_stat *ts)
2170 if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0)
2176 * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim
2178 * @td: FIO thread data.
2179 * @io_u: FIO I/O unit.
2181 * It is assumed that z->mutex is already locked.
2182 * Return io_u_completed when reset zone succeeds. Return 0 when the target zone
2183 * does not have write pointer. On error, return negative errno.
2185 int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u)
2187 struct fio_file *f = io_u->file;
2188 struct fio_zone_info *z;
2191 z = zbd_offset_to_zone(f, io_u->offset);
2195 if (io_u->offset != z->start) {
2196 log_err("Trim offset not at zone start (%lld)\n",
2201 ret = zbd_reset_zone((struct thread_data *)td, f, z);
2205 return io_u_completed;