2 * Copyright (C) 2018 Western Digital Corporation or its affiliates.
4 * This file is released under the GPL.
14 #include "compiler/compiler.h"
20 #include "oslib/asprintf.h"
26 static bool is_valid_offset(const struct fio_file *f, uint64_t offset)
28 return (uint64_t)(offset - f->file_offset) < f->io_size;
31 static inline unsigned int zbd_zone_idx(const struct fio_file *f,
32 struct fio_zone_info *zone)
34 return zone - f->zbd_info->zone_info;
38 * zbd_offset_to_zone_idx - convert an offset into a zone number
40 * @offset: offset in bytes. If this offset is in the first zone_size bytes
41 * past the disk size then the index of the sentinel is returned.
43 static unsigned int zbd_offset_to_zone_idx(const struct fio_file *f,
48 if (f->zbd_info->zone_size_log2 > 0)
49 zone_idx = offset >> f->zbd_info->zone_size_log2;
51 zone_idx = offset / f->zbd_info->zone_size;
53 return min(zone_idx, f->zbd_info->nr_zones);
57 * zbd_zone_end - Return zone end location
58 * @z: zone info pointer.
60 static inline uint64_t zbd_zone_end(const struct fio_zone_info *z)
66 * zbd_zone_capacity_end - Return zone capacity limit end location
67 * @z: zone info pointer.
69 static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z)
71 return z->start + z->capacity;
75 * zbd_zone_remainder - Return the number of bytes that are still available for
76 * writing before the zone gets full
77 * @z: zone info pointer.
79 static inline uint64_t zbd_zone_remainder(struct fio_zone_info *z)
81 if (z->wp >= zbd_zone_capacity_end(z))
84 return zbd_zone_capacity_end(z) - z->wp;
88 * zbd_zone_full - verify whether a minimum number of bytes remain in a zone
90 * @z: zone info pointer.
91 * @required: minimum number of bytes that must remain in a zone.
93 * The caller must hold z->mutex.
95 static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z,
98 assert((required & 511) == 0);
100 return z->has_wp && required > zbd_zone_remainder(z);
103 static void zone_lock(struct thread_data *td, const struct fio_file *f,
104 struct fio_zone_info *z)
107 struct zoned_block_device_info *zbd = f->zbd_info;
108 uint32_t const nz = z - zbd->zone_info;
109 /* A thread should never lock zones outside its working area. */
110 assert(f->min_zone <= nz && nz < f->max_zone);
115 * Lock the io_u target zone. The zone will be unlocked if io_u offset
116 * is changed or when io_u completes and zbd_put_io() executed.
117 * To avoid multiple jobs doing asynchronous I/Os from deadlocking each
118 * other waiting for zone locks when building an io_u batch, first
119 * only trylock the zone. If the zone is already locked by another job,
120 * process the currently queued I/Os so that I/O progress is made and
123 if (pthread_mutex_trylock(&z->mutex) != 0) {
124 if (!td_ioengine_flagged(td, FIO_SYNCIO))
126 pthread_mutex_lock(&z->mutex);
130 static inline void zone_unlock(struct fio_zone_info *z)
133 pthread_mutex_unlock(&z->mutex);
136 static inline struct fio_zone_info *zbd_get_zone(const struct fio_file *f,
137 unsigned int zone_idx)
139 return &f->zbd_info->zone_info[zone_idx];
142 static inline struct fio_zone_info *
143 zbd_offset_to_zone(const struct fio_file *f, uint64_t offset)
145 return zbd_get_zone(f, zbd_offset_to_zone_idx(f, offset));
148 static bool accounting_vdb(struct thread_data *td, const struct fio_file *f)
150 return td->o.zrt.u.f && td_write(td);
154 * zbd_get_zoned_model - Get a device zoned model
155 * @td: FIO thread data
156 * @f: FIO file for which to get model information
158 static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f,
159 enum zbd_zoned_model *model)
163 if (f->filetype == FIO_TYPE_PIPE) {
164 log_err("zonemode=zbd does not support pipes\n");
168 /* If regular file, always emulate zones inside the file. */
169 if (f->filetype == FIO_TYPE_FILE) {
174 if (td->io_ops && td->io_ops->get_zoned_model)
175 ret = td->io_ops->get_zoned_model(td, f, model);
177 ret = blkzoned_get_zoned_model(td, f, model);
179 td_verror(td, errno, "get zoned model failed");
180 log_err("%s: get zoned model failed (%d).\n",
181 f->file_name, errno);
188 * zbd_report_zones - Get zone information
189 * @td: FIO thread data.
190 * @f: FIO file for which to get zone information
191 * @offset: offset from which to report zones
192 * @zones: Array of struct zbd_zone
193 * @nr_zones: Size of @zones array
195 * Get zone information into @zones starting from the zone at offset @offset
196 * for the device specified by @f.
198 * Returns the number of zones reported upon success and a negative error code
199 * upon failure. If the zone report is empty, always assume an error (device
200 * problem) and return -EIO.
202 static int zbd_report_zones(struct thread_data *td, struct fio_file *f,
203 uint64_t offset, struct zbd_zone *zones,
204 unsigned int nr_zones)
208 if (td->io_ops && td->io_ops->report_zones)
209 ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones);
211 ret = blkzoned_report_zones(td, f, offset, zones, nr_zones);
213 td_verror(td, errno, "report zones failed");
214 log_err("%s: report zones from sector %"PRIu64" failed (nr_zones=%d; errno=%d).\n",
215 f->file_name, offset >> 9, nr_zones, errno);
216 } else if (ret == 0) {
217 td_verror(td, errno, "Empty zone report");
218 log_err("%s: report zones from sector %"PRIu64" is empty.\n",
219 f->file_name, offset >> 9);
227 * zbd_reset_wp - reset the write pointer of a range of zones
228 * @td: FIO thread data.
229 * @f: FIO file for which to reset zones
230 * @offset: Starting offset of the first zone to reset
231 * @length: Length of the range of zones to reset
233 * Reset the write pointer of all zones in the range @offset...@offset+@length.
234 * Returns 0 upon success and a negative error code upon failure.
236 static int zbd_reset_wp(struct thread_data *td, struct fio_file *f,
237 uint64_t offset, uint64_t length)
241 if (td->io_ops && td->io_ops->reset_wp)
242 ret = td->io_ops->reset_wp(td, f, offset, length);
244 ret = blkzoned_reset_wp(td, f, offset, length);
246 td_verror(td, errno, "resetting wp failed");
247 log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n",
248 f->file_name, length >> 9, offset >> 9, errno);
255 * __zbd_reset_zone - reset the write pointer of a single zone
256 * @td: FIO thread data.
257 * @f: FIO file associated with the disk for which to reset a write pointer.
260 * Returns 0 upon success and a negative error code upon failure.
262 * The caller must hold z->mutex.
264 static int __zbd_reset_zone(struct thread_data *td, struct fio_file *f,
265 struct fio_zone_info *z)
267 uint64_t offset = z->start;
268 uint64_t length = (z+1)->start - offset;
269 uint64_t data_in_zone = z->wp - z->start;
275 assert(is_valid_offset(f, offset + length - 1));
277 dprint(FD_ZBD, "%s: resetting wp of zone %u.\n",
278 f->file_name, zbd_zone_idx(f, z));
280 switch (f->zbd_info->model) {
282 case ZBD_HOST_MANAGED:
283 ret = zbd_reset_wp(td, f, offset, length);
291 if (accounting_vdb(td, f)) {
292 pthread_mutex_lock(&f->zbd_info->mutex);
293 f->zbd_info->wp_valid_data_bytes -= data_in_zone;
294 pthread_mutex_unlock(&f->zbd_info->mutex);
299 td->ts.nr_zone_resets++;
305 * zbd_write_zone_put - Remove a zone from the write target zones array.
306 * @td: FIO thread data.
307 * @f: FIO file that has the write zones array to remove.
308 * @zone_idx: Index of the zone to remove.
310 * The caller must hold f->zbd_info->mutex.
312 static void zbd_write_zone_put(struct thread_data *td, const struct fio_file *f,
313 struct fio_zone_info *z)
320 for (zi = 0; zi < f->zbd_info->num_write_zones; zi++) {
321 if (zbd_get_zone(f, f->zbd_info->write_zones[zi]) == z)
324 if (zi == f->zbd_info->num_write_zones)
327 dprint(FD_ZBD, "%s: removing zone %u from write zone array\n",
328 f->file_name, zbd_zone_idx(f, z));
330 memmove(f->zbd_info->write_zones + zi,
331 f->zbd_info->write_zones + zi + 1,
332 (ZBD_MAX_WRITE_ZONES - (zi + 1)) *
333 sizeof(f->zbd_info->write_zones[0]));
335 f->zbd_info->num_write_zones--;
336 td->num_write_zones--;
341 * zbd_reset_zone - reset the write pointer of a single zone and remove the zone
342 * from the array of write zones.
343 * @td: FIO thread data.
344 * @f: FIO file associated with the disk for which to reset a write pointer.
347 * Returns 0 upon success and a negative error code upon failure.
349 * The caller must hold z->mutex.
351 static int zbd_reset_zone(struct thread_data *td, struct fio_file *f,
352 struct fio_zone_info *z)
356 ret = __zbd_reset_zone(td, f, z);
360 pthread_mutex_lock(&f->zbd_info->mutex);
361 zbd_write_zone_put(td, f, z);
362 pthread_mutex_unlock(&f->zbd_info->mutex);
367 * zbd_finish_zone - finish the specified zone
368 * @td: FIO thread data.
369 * @f: FIO file for which to finish a zone
370 * @z: Zone to finish.
372 * Finish the zone at @offset with open or close status.
374 static int zbd_finish_zone(struct thread_data *td, struct fio_file *f,
375 struct fio_zone_info *z)
377 uint64_t offset = z->start;
378 uint64_t length = f->zbd_info->zone_size;
381 switch (f->zbd_info->model) {
383 case ZBD_HOST_MANAGED:
384 if (td->io_ops && td->io_ops->finish_zone)
385 ret = td->io_ops->finish_zone(td, f, offset, length);
387 ret = blkzoned_finish_zone(td, f, offset, length);
394 td_verror(td, errno, "finish zone failed");
395 log_err("%s: finish zone at sector %"PRIu64" failed (%d).\n",
396 f->file_name, offset >> 9, errno);
398 z->wp = (z+1)->start;
405 * zbd_reset_zones - Reset a range of zones.
406 * @td: fio thread data.
407 * @f: fio file for which to reset zones
408 * @zb: first zone to reset.
409 * @ze: first zone not to reset.
411 * Returns 0 upon success and 1 upon failure.
413 static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
414 struct fio_zone_info *const zb,
415 struct fio_zone_info *const ze)
417 struct fio_zone_info *z;
418 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
421 if (fio_unlikely(0 == min_bs))
424 dprint(FD_ZBD, "%s: examining zones %u .. %u\n",
425 f->file_name, zbd_zone_idx(f, zb), zbd_zone_idx(f, ze));
427 for (z = zb; z < ze; z++) {
433 if (z->wp != z->start) {
434 dprint(FD_ZBD, "%s: resetting zone %u\n",
435 f->file_name, zbd_zone_idx(f, z));
436 if (zbd_reset_zone(td, f, z) < 0)
447 * zbd_get_max_open_zones - Get the maximum number of open zones
448 * @td: FIO thread data
449 * @f: FIO file for which to get max open zones
450 * @max_open_zones: Upon success, result will be stored here.
452 * A @max_open_zones value set to zero means no limit.
454 * Returns 0 upon success and a negative error code upon failure.
456 static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f,
457 unsigned int *max_open_zones)
461 if (td->io_ops && td->io_ops->get_max_open_zones)
462 ret = td->io_ops->get_max_open_zones(td, f, max_open_zones);
464 ret = blkzoned_get_max_open_zones(td, f, max_open_zones);
466 td_verror(td, errno, "get max open zones failed");
467 log_err("%s: get max open zones failed (%d).\n",
468 f->file_name, errno);
475 * zbd_get_max_active_zones - Get the maximum number of active zones
476 * @td: FIO thread data
477 * @f: FIO file for which to get max active zones
479 * Returns max_active_zones limit value of the target file if it is available.
480 * Otherwise return zero, which means no limit.
482 static unsigned int zbd_get_max_active_zones(struct thread_data *td,
485 unsigned int max_active_zones;
488 if (td->io_ops && td->io_ops->get_max_active_zones)
489 ret = td->io_ops->get_max_active_zones(td, f,
492 ret = blkzoned_get_max_active_zones(td, f, &max_active_zones);
494 dprint(FD_ZBD, "%s: max_active_zones is not available\n",
499 return max_active_zones;
503 * __zbd_write_zone_get - Add a zone to the array of write zones.
504 * @td: fio thread data.
505 * @f: fio file that has the write zones array to add.
506 * @zone_idx: Index of the zone to add.
508 * Do same operation as @zbd_write_zone_get, except it adds the zone at
509 * @zone_idx to write target zones array even when it does not have remainder
510 * space to write one block.
512 static bool __zbd_write_zone_get(struct thread_data *td,
513 const struct fio_file *f,
514 struct fio_zone_info *z)
516 struct zoned_block_device_info *zbdi = f->zbd_info;
517 uint32_t zone_idx = zbd_zone_idx(f, z);
520 if (z->cond == ZBD_ZONE_COND_OFFLINE)
524 * Skip full zones with data verification enabled because resetting a
525 * zone causes data loss and hence causes verification to fail.
527 if (td->o.verify != VERIFY_NONE && zbd_zone_remainder(z) == 0)
531 * zbdi->max_write_zones == 0 means that there is no limit on the
532 * maximum number of write target zones. In this case, do no track write
533 * target zones in zbdi->write_zones array.
535 if (!zbdi->max_write_zones)
538 pthread_mutex_lock(&zbdi->mutex);
542 * If the zone is going to be completely filled by writes
543 * already in-flight, handle it as a full zone instead of a
546 if (!zbd_zone_remainder(z))
552 /* Zero means no limit */
553 if (td->o.job_max_open_zones > 0 &&
554 td->num_write_zones >= td->o.job_max_open_zones)
556 if (zbdi->num_write_zones >= zbdi->max_write_zones)
559 dprint(FD_ZBD, "%s: adding zone %u to write zone array\n",
560 f->file_name, zone_idx);
562 zbdi->write_zones[zbdi->num_write_zones++] = zone_idx;
563 td->num_write_zones++;
568 pthread_mutex_unlock(&zbdi->mutex);
573 * zbd_write_zone_get - Add a zone to the array of write zones.
574 * @td: fio thread data.
575 * @f: fio file that has the open zones to add.
576 * @zone_idx: Index of the zone to add.
578 * Add a ZBD zone to write target zones array, if it is not yet added. Returns
579 * true if either the zone was already added or if the zone was successfully
580 * added to the array without exceeding the maximum number of write zones.
581 * Returns false if the zone was not already added and addition of the zone
582 * would cause the zone limit to be exceeded.
584 static bool zbd_write_zone_get(struct thread_data *td, const struct fio_file *f,
585 struct fio_zone_info *z)
587 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
590 * Skip full zones with data verification enabled because resetting a
591 * zone causes data loss and hence causes verification to fail.
593 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
596 return __zbd_write_zone_get(td, f, z);
599 /* Verify whether direct I/O is used for all host-managed zoned block drives. */
600 static bool zbd_using_direct_io(void)
606 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
608 for_each_file(td, f, j) {
609 if (f->zbd_info && f->filetype == FIO_TYPE_BLOCK &&
610 f->zbd_info->model == ZBD_HOST_MANAGED)
618 /* Whether or not the I/O range for f includes one or more sequential zones */
619 static bool zbd_is_seq_job(const struct fio_file *f)
621 uint32_t zone_idx, zone_idx_b, zone_idx_e;
628 zone_idx_b = zbd_offset_to_zone_idx(f, f->file_offset);
630 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size - 1);
631 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
632 if (zbd_get_zone(f, zone_idx)->has_wp)
639 * Verify whether the file offset and size parameters are aligned with zone
640 * boundaries. If the file offset is not aligned, align it down to the start of
641 * the zone containing the start offset and align up the file io_size parameter.
643 static bool zbd_zone_align_file_sizes(struct thread_data *td,
646 const struct fio_zone_info *z;
647 uint64_t new_offset, new_end;
651 if (f->file_offset >= f->real_file_size)
653 if (!zbd_is_seq_job(f))
656 if (!td->o.zone_size) {
657 td->o.zone_size = f->zbd_info->zone_size;
658 if (!td->o.zone_size) {
659 log_err("%s: invalid 0 zone size\n",
663 } else if (td->o.zone_size != f->zbd_info->zone_size) {
664 log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n",
665 f->file_name, td->o.zone_size,
666 f->zbd_info->zone_size);
670 if (td->o.zone_skip % td->o.zone_size) {
671 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
672 f->file_name, td->o.zone_skip,
677 z = zbd_offset_to_zone(f, f->file_offset);
678 if ((f->file_offset != z->start) &&
679 (td->o.td_ddir != TD_DDIR_READ)) {
680 new_offset = zbd_zone_end(z);
681 if (new_offset >= f->file_offset + f->io_size) {
682 log_info("%s: io_size must be at least one zone\n",
686 log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n",
687 f->file_name, f->file_offset,
689 f->io_size -= (new_offset - f->file_offset);
690 f->file_offset = new_offset;
693 z = zbd_offset_to_zone(f, f->file_offset + f->io_size);
695 if ((td->o.td_ddir != TD_DDIR_READ) &&
696 (f->file_offset + f->io_size != new_end)) {
697 if (new_end <= f->file_offset) {
698 log_info("%s: io_size must be at least one zone\n",
702 log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n",
703 f->file_name, f->io_size,
704 new_end - f->file_offset);
705 f->io_size = new_end - f->file_offset;
712 * Verify whether offset and size parameters are aligned with zone boundaries.
714 static bool zbd_verify_sizes(void)
720 for_each_file(td, f, j) {
721 if (!zbd_zone_align_file_sizes(td, f))
729 static bool zbd_verify_bs(void)
736 (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] ||
737 td->o.bssplit_nr[DDIR_TRIM])) {
738 log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n");
741 for_each_file(td, f, j) {
747 zone_size = f->zbd_info->zone_size;
748 if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) {
749 log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n",
750 f->file_name, td->o.bs[DDIR_TRIM],
759 static int ilog2(uint64_t i)
771 * Initialize f->zbd_info for devices that are not zoned block devices. This
772 * allows to execute a ZBD workload against a non-ZBD device.
774 static int init_zone_info(struct thread_data *td, struct fio_file *f)
777 struct fio_zone_info *p;
778 uint64_t zone_size = td->o.zone_size;
779 uint64_t zone_capacity = td->o.zone_capacity;
780 struct zoned_block_device_info *zbd_info = NULL;
783 if (zone_size == 0) {
784 log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n",
789 if (zone_size < 512) {
790 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
795 if (zone_capacity == 0)
796 zone_capacity = zone_size;
798 if (zone_capacity > zone_size) {
799 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
800 f->file_name, td->o.zone_capacity, td->o.zone_size);
804 if (f->real_file_size < zone_size) {
805 log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n",
806 f->file_name, f->real_file_size, zone_size);
810 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
811 zbd_info = scalloc(1, sizeof(*zbd_info) +
812 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
816 mutex_init_pshared(&zbd_info->mutex);
817 zbd_info->refcount = 1;
818 p = &zbd_info->zone_info[0];
819 for (i = 0; i < nr_zones; i++, p++) {
820 mutex_init_pshared_with_type(&p->mutex,
821 PTHREAD_MUTEX_RECURSIVE);
822 p->start = i * zone_size;
824 p->type = ZBD_ZONE_TYPE_SWR;
825 p->cond = ZBD_ZONE_COND_EMPTY;
826 p->capacity = zone_capacity;
830 p->start = nr_zones * zone_size;
832 f->zbd_info = zbd_info;
833 f->zbd_info->zone_size = zone_size;
834 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
835 ilog2(zone_size) : 0;
836 f->zbd_info->nr_zones = nr_zones;
841 * Maximum number of zones to report in one operation.
843 #define ZBD_REPORT_MAX_ZONES 8192U
846 * Parse the device zone report and store it in f->zbd_info. Must be called
847 * only for devices that are zoned, namely those with a model != ZBD_NONE.
849 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
852 struct zbd_zone *zones, *z;
853 struct fio_zone_info *p;
854 uint64_t zone_size, offset, capacity;
855 bool same_zone_cap = true;
856 struct zoned_block_device_info *zbd_info = NULL;
857 int i, j, ret = -ENOMEM;
859 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
863 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
866 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
871 zone_size = zones[0].len;
872 capacity = zones[0].capacity;
873 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
875 if (td->o.zone_size == 0) {
876 td->o.zone_size = zone_size;
877 } else if (td->o.zone_size != zone_size) {
878 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
879 f->file_name, td->o.zone_size, zone_size);
884 dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n",
885 f->file_name, nr_zones, zone_size / 1024);
887 zbd_info = scalloc(1, sizeof(*zbd_info) +
888 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
891 mutex_init_pshared(&zbd_info->mutex);
892 zbd_info->refcount = 1;
893 p = &zbd_info->zone_info[0];
894 for (offset = 0, j = 0; j < nr_zones;) {
896 for (i = 0; i < nrz; i++, j++, z++, p++) {
897 mutex_init_pshared_with_type(&p->mutex,
898 PTHREAD_MUTEX_RECURSIVE);
900 p->capacity = z->capacity;
901 if (capacity != z->capacity)
902 same_zone_cap = false;
905 case ZBD_ZONE_COND_NOT_WP:
906 case ZBD_ZONE_COND_FULL:
907 p->wp = p->start + p->capacity;
910 assert(z->start <= z->wp);
911 assert(z->wp <= z->start + zone_size);
917 case ZBD_ZONE_TYPE_SWR:
926 if (j > 0 && p->start != p[-1].start + zone_size) {
927 log_info("%s: invalid zone data [%d:%d]: %"PRIu64" + %"PRIu64" != %"PRIu64"\n",
929 p[-1].start, zone_size, p->start);
935 offset = z->start + z->len;
939 nrz = zbd_report_zones(td, f, offset, zones,
940 min((uint32_t)(nr_zones - j),
941 ZBD_REPORT_MAX_ZONES));
944 log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n",
945 offset, f->file_name, -ret);
951 zbd_info->zone_info[nr_zones].start = offset;
953 f->zbd_info = zbd_info;
954 f->zbd_info->zone_size = zone_size;
955 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
956 ilog2(zone_size) : 0;
957 f->zbd_info->nr_zones = nr_zones;
958 f->zbd_info->max_active_zones = zbd_get_max_active_zones(td, f);
961 dprint(FD_ZBD, "Zone capacity = %"PRIu64" KB\n",
973 static int zbd_set_max_write_zones(struct thread_data *td, struct fio_file *f)
975 struct zoned_block_device_info *zbd = f->zbd_info;
976 unsigned int max_open_zones;
979 if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) {
980 /* Only host-managed devices have a max open limit */
981 zbd->max_write_zones = td->o.max_open_zones;
985 /* If host-managed, get the max open limit */
986 ret = zbd_get_max_open_zones(td, f, &max_open_zones);
990 if (!max_open_zones) {
991 /* No device limit */
992 zbd->max_write_zones = td->o.max_open_zones;
993 } else if (!td->o.max_open_zones) {
994 /* No user limit. Set limit to device limit */
995 zbd->max_write_zones = max_open_zones;
996 } else if (td->o.max_open_zones <= max_open_zones) {
997 /* Both user limit and dev limit. User limit not too large */
998 zbd->max_write_zones = td->o.max_open_zones;
1000 /* Both user limit and dev limit. User limit too large */
1001 td_verror(td, EINVAL,
1002 "Specified --max_open_zones is too large");
1003 log_err("Specified --max_open_zones (%d) is larger than max (%u)\n",
1004 td->o.max_open_zones, max_open_zones);
1009 /* Ensure that the limit is not larger than FIO's internal limit */
1010 if (zbd->max_write_zones > ZBD_MAX_WRITE_ZONES) {
1011 td_verror(td, EINVAL, "'max_open_zones' value is too large");
1012 log_err("'max_open_zones' value is larger than %u\n",
1013 ZBD_MAX_WRITE_ZONES);
1017 dprint(FD_ZBD, "%s: using max write zones limit: %"PRIu32"\n",
1018 f->file_name, zbd->max_write_zones);
1024 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
1026 * Returns 0 upon success and a negative error code upon failure.
1028 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
1030 enum zbd_zoned_model zbd_model;
1033 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1035 ret = zbd_get_zoned_model(td, f, &zbd_model);
1039 switch (zbd_model) {
1040 case ZBD_HOST_AWARE:
1041 case ZBD_HOST_MANAGED:
1042 ret = parse_zone_info(td, f);
1047 ret = init_zone_info(td, f);
1052 td_verror(td, EINVAL, "Unsupported zoned model");
1053 log_err("Unsupported zoned model\n");
1057 assert(f->zbd_info);
1058 f->zbd_info->model = zbd_model;
1060 ret = zbd_set_max_write_zones(td, f);
1062 zbd_free_zone_info(f);
1069 void zbd_free_zone_info(struct fio_file *f)
1073 assert(f->zbd_info);
1075 pthread_mutex_lock(&f->zbd_info->mutex);
1076 refcount = --f->zbd_info->refcount;
1077 pthread_mutex_unlock(&f->zbd_info->mutex);
1079 assert((int32_t)refcount >= 0);
1086 * Initialize f->zbd_info.
1088 * Returns 0 upon success and a negative error code upon failure.
1090 * Note: this function can only work correctly if it is called before the first
1093 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
1095 struct fio_file *f2;
1099 for_each_file(td2, f2, j) {
1100 if (td2 == td && f2 == file)
1102 if (!f2->zbd_info ||
1103 strcmp(f2->file_name, file->file_name) != 0)
1105 file->zbd_info = f2->zbd_info;
1106 file->zbd_info->refcount++;
1111 ret = zbd_create_zone_info(td, file);
1113 td_verror(td, -ret, "zbd_create_zone_info() failed");
1118 int zbd_init_files(struct thread_data *td)
1123 for_each_file(td, f, i) {
1124 if (zbd_init_zone_info(td, f))
1131 void zbd_recalc_options_with_zone_granularity(struct thread_data *td)
1136 for_each_file(td, f, i) {
1137 struct zoned_block_device_info *zbd = f->zbd_info;
1140 /* zonemode=strided doesn't get per-file zone size. */
1141 zone_size = zbd ? zbd->zone_size : td->o.zone_size;
1145 if (td->o.size_nz > 0)
1146 td->o.size = td->o.size_nz * zone_size;
1147 if (td->o.io_size_nz > 0)
1148 td->o.io_size = td->o.io_size_nz * zone_size;
1149 if (td->o.start_offset_nz > 0)
1150 td->o.start_offset = td->o.start_offset_nz * zone_size;
1151 if (td->o.offset_increment_nz > 0)
1152 td->o.offset_increment =
1153 td->o.offset_increment_nz * zone_size;
1154 if (td->o.zone_skip_nz > 0)
1155 td->o.zone_skip = td->o.zone_skip_nz * zone_size;
1159 static uint64_t zbd_verify_and_set_vdb(struct thread_data *td,
1160 const struct fio_file *f)
1162 struct fio_zone_info *zb, *ze, *z;
1163 uint64_t wp_vdb = 0;
1164 struct zoned_block_device_info *zbdi = f->zbd_info;
1166 assert(td->runstate < TD_RUNNING);
1169 if (!accounting_vdb(td, f))
1173 * Ensure that the I/O range includes one or more sequential zones so
1174 * that f->min_zone and f->max_zone have different values.
1176 if (!zbd_is_seq_job(f))
1179 if (zbdi->write_min_zone != zbdi->write_max_zone) {
1180 if (zbdi->write_min_zone != f->min_zone ||
1181 zbdi->write_max_zone != f->max_zone) {
1182 td_verror(td, EINVAL,
1183 "multi-jobs with different write ranges are "
1184 "not supported with zone_reset_threshold");
1185 log_err("multi-jobs with different write ranges are "
1186 "not supported with zone_reset_threshold\n");
1191 zbdi->write_min_zone = f->min_zone;
1192 zbdi->write_max_zone = f->max_zone;
1194 zb = zbd_get_zone(f, f->min_zone);
1195 ze = zbd_get_zone(f, f->max_zone);
1196 for (z = zb; z < ze; z++)
1198 wp_vdb += z->wp - z->start;
1200 zbdi->wp_valid_data_bytes = wp_vdb;
1205 int zbd_setup_files(struct thread_data *td)
1210 if (!zbd_using_direct_io()) {
1211 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
1215 if (!zbd_verify_sizes())
1218 if (!zbd_verify_bs())
1221 if (td->o.experimental_verify) {
1222 log_err("zonemode=zbd does not support experimental verify\n");
1226 for_each_file(td, f, i) {
1227 struct zoned_block_device_info *zbd = f->zbd_info;
1228 struct fio_zone_info *z;
1234 f->min_zone = zbd_offset_to_zone_idx(f, f->file_offset);
1236 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size);
1238 vdb = zbd_verify_and_set_vdb(td, f);
1240 dprint(FD_ZBD, "%s(%s): valid data bytes = %" PRIu64 "\n",
1241 __func__, f->file_name, vdb);
1244 * When all zones in the I/O range are conventional, io_size
1245 * can be smaller than zone size, making min_zone the same
1246 * as max_zone. This is why the assert below needs to be made
1249 if (zbd_is_seq_job(f))
1250 assert(f->min_zone < f->max_zone);
1252 if (td->o.max_open_zones > 0 &&
1253 zbd->max_write_zones != td->o.max_open_zones) {
1254 log_err("Different 'max_open_zones' values\n");
1259 * The per job max open zones limit cannot be used without a
1260 * global max open zones limit. (As the tracking of open zones
1261 * is disabled when there is no global max open zones limit.)
1263 if (td->o.job_max_open_zones && !zbd->max_write_zones) {
1264 log_err("'job_max_open_zones' cannot be used without a global open zones limit\n");
1269 * zbd->max_write_zones is the global limit shared for all jobs
1270 * that target the same zoned block device. Force sync the per
1271 * thread global limit with the actual global limit. (The real
1272 * per thread/job limit is stored in td->o.job_max_open_zones).
1274 td->o.max_open_zones = zbd->max_write_zones;
1276 for (zi = f->min_zone; zi < f->max_zone; zi++) {
1277 z = &zbd->zone_info[zi];
1278 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
1279 z->cond != ZBD_ZONE_COND_EXP_OPEN &&
1280 z->cond != ZBD_ZONE_COND_CLOSED)
1282 if (!zbd->max_active_zones &&
1283 z->cond == ZBD_ZONE_COND_CLOSED)
1285 if (__zbd_write_zone_get(td, f, z))
1288 * If the number of open zones exceeds specified limits,
1291 log_err("Number of open zones exceeds max_open_zones limit\n");
1300 * Reset zbd_info.write_cnt, the counter that counts down towards the next
1303 static void _zbd_reset_write_cnt(const struct thread_data *td,
1304 const struct fio_file *f)
1306 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
1308 f->zbd_info->write_cnt = td->o.zrf.u.f ?
1309 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
1312 static void zbd_reset_write_cnt(const struct thread_data *td,
1313 const struct fio_file *f)
1315 pthread_mutex_lock(&f->zbd_info->mutex);
1316 _zbd_reset_write_cnt(td, f);
1317 pthread_mutex_unlock(&f->zbd_info->mutex);
1320 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
1321 const struct fio_file *f)
1323 uint32_t write_cnt = 0;
1325 pthread_mutex_lock(&f->zbd_info->mutex);
1326 assert(f->zbd_info->write_cnt);
1327 if (f->zbd_info->write_cnt)
1328 write_cnt = --f->zbd_info->write_cnt;
1330 _zbd_reset_write_cnt(td, f);
1331 pthread_mutex_unlock(&f->zbd_info->mutex);
1333 return write_cnt == 0;
1336 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
1338 struct fio_zone_info *zb, *ze;
1339 bool verify_data_left = false;
1341 if (!f->zbd_info || !td_write(td))
1344 zb = zbd_get_zone(f, f->min_zone);
1345 ze = zbd_get_zone(f, f->max_zone);
1348 * If data verification is enabled reset the affected zones before
1349 * writing any data to avoid that a zone reset has to be issued while
1350 * writing data, which causes data loss.
1352 if (td->o.verify != VERIFY_NONE) {
1353 verify_data_left = td->runstate == TD_VERIFYING ||
1354 td->io_hist_len || td->verify_batch;
1355 if (td->io_hist_len && td->o.verify_backlog)
1357 td->io_hist_len % td->o.verify_backlog;
1358 if (!verify_data_left)
1359 zbd_reset_zones(td, f, zb, ze);
1362 zbd_reset_write_cnt(td, f);
1365 /* Return random zone index for one of the write target zones. */
1366 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1367 const struct io_u *io_u)
1369 return (io_u->offset - f->file_offset) *
1370 f->zbd_info->num_write_zones / f->io_size;
1373 static bool any_io_in_flight(void)
1376 if (td->io_u_in_flight)
1384 * Modify the offset of an I/O unit that does not refer to a zone such that
1385 * in write target zones array. Add a zone to or remove a zone from the lsit if
1386 * necessary. The write target zone is searched across sequential zones.
1387 * This algorithm can only work correctly if all write pointers are
1388 * a multiple of the fio block size. The caller must neither hold z->mutex
1389 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1391 static struct fio_zone_info *zbd_convert_to_write_zone(struct thread_data *td,
1394 const uint64_t min_bs = td->o.min_bs[io_u->ddir];
1395 struct fio_file *f = io_u->file;
1396 struct zoned_block_device_info *zbdi = f->zbd_info;
1397 struct fio_zone_info *z;
1398 unsigned int write_zone_idx = -1;
1399 uint32_t zone_idx, new_zone_idx;
1401 bool wait_zone_write;
1403 bool should_retry = true;
1405 assert(is_valid_offset(f, io_u->offset));
1407 if (zbdi->max_write_zones || td->o.job_max_open_zones) {
1409 * This statement accesses zbdi->write_zones[] on purpose
1412 zone_idx = zbdi->write_zones[pick_random_zone_idx(f, io_u)];
1414 zone_idx = zbd_offset_to_zone_idx(f, io_u->offset);
1416 if (zone_idx < f->min_zone)
1417 zone_idx = f->min_zone;
1418 else if (zone_idx >= f->max_zone)
1419 zone_idx = f->max_zone - 1;
1422 "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1423 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1426 * Since z->mutex is the outer lock and zbdi->mutex the inner
1427 * lock it can happen that the state of the zone with index zone_idx
1428 * has changed after 'z' has been assigned and before zbdi->mutex
1429 * has been obtained. Hence the loop.
1434 z = zbd_get_zone(f, zone_idx);
1436 zone_lock(td, f, z);
1438 pthread_mutex_lock(&zbdi->mutex);
1441 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1442 zbdi->max_write_zones == 0 &&
1443 td->o.job_max_open_zones == 0)
1445 if (zbdi->num_write_zones == 0) {
1446 dprint(FD_ZBD, "%s(%s): no zone is write target\n",
1447 __func__, f->file_name);
1448 goto choose_other_zone;
1453 * Array of write target zones is per-device, shared across all
1454 * threads. Start with quasi-random candidate zone. Ignore
1455 * zones which don't belong to thread's offset/size area.
1457 write_zone_idx = pick_random_zone_idx(f, io_u);
1458 assert(!write_zone_idx ||
1459 write_zone_idx < zbdi->num_write_zones);
1460 tmp_idx = write_zone_idx;
1462 for (i = 0; i < zbdi->num_write_zones; i++) {
1465 if (tmp_idx >= zbdi->num_write_zones)
1467 tmpz = zbdi->write_zones[tmp_idx];
1468 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1469 write_zone_idx = tmp_idx;
1470 goto found_candidate_zone;
1476 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1477 __func__, f->file_name);
1479 pthread_mutex_unlock(&zbdi->mutex);
1486 found_candidate_zone:
1487 new_zone_idx = zbdi->write_zones[write_zone_idx];
1488 if (new_zone_idx == zone_idx)
1490 zone_idx = new_zone_idx;
1492 pthread_mutex_unlock(&zbdi->mutex);
1498 /* Both z->mutex and zbdi->mutex are held. */
1501 if (zbd_zone_remainder(z) >= min_bs) {
1502 pthread_mutex_unlock(&zbdi->mutex);
1507 /* Check if number of write target zones reaches one of limits. */
1509 zbdi->num_write_zones == f->max_zone - f->min_zone ||
1510 (zbdi->max_write_zones &&
1511 zbdi->num_write_zones == zbdi->max_write_zones) ||
1512 (td->o.job_max_open_zones &&
1513 td->num_write_zones == td->o.job_max_open_zones);
1515 pthread_mutex_unlock(&zbdi->mutex);
1517 /* Only z->mutex is held. */
1520 * When number of write target zones reaches to one of limits, wait for
1521 * zone write completion to one of them before trying a new zone.
1523 if (wait_zone_write) {
1525 "%s(%s): quiesce to remove a zone from write target zones array\n",
1526 __func__, f->file_name);
1531 /* Zone 'z' is full, so try to choose a new zone. */
1532 for (i = f->io_size / zbdi->zone_size; i > 0; i--) {
1537 if (!is_valid_offset(f, z->start)) {
1539 zone_idx = f->min_zone;
1540 z = zbd_get_zone(f, zone_idx);
1542 assert(is_valid_offset(f, z->start));
1545 zone_lock(td, f, z);
1548 if (zbd_write_zone_get(td, f, z))
1552 /* Only z->mutex is held. */
1554 /* Check whether the write fits in any of the write target zones. */
1555 pthread_mutex_lock(&zbdi->mutex);
1556 for (i = 0; i < zbdi->num_write_zones; i++) {
1557 zone_idx = zbdi->write_zones[i];
1558 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1560 pthread_mutex_unlock(&zbdi->mutex);
1563 z = zbd_get_zone(f, zone_idx);
1565 zone_lock(td, f, z);
1566 if (zbd_zone_remainder(z) >= min_bs)
1568 pthread_mutex_lock(&zbdi->mutex);
1572 * When any I/O is in-flight or when all I/Os in-flight get completed,
1573 * the I/Os might have removed zones from the write target array then
1574 * retry the steps to choose a zone. Before retry, call io_u_quiesce()
1575 * to complete in-flight writes.
1577 in_flight = any_io_in_flight();
1578 if (in_flight || should_retry) {
1580 "%s(%s): wait zone write and retry write target zone selection\n",
1581 __func__, f->file_name);
1582 should_retry = in_flight;
1583 pthread_mutex_unlock(&zbdi->mutex);
1586 zone_lock(td, f, z);
1590 pthread_mutex_unlock(&zbdi->mutex);
1594 dprint(FD_ZBD, "%s(%s): did not choose another write zone\n",
1595 __func__, f->file_name);
1600 dprint(FD_ZBD, "%s(%s): returning zone %d\n",
1601 __func__, f->file_name, zone_idx);
1603 io_u->offset = z->start;
1605 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1611 * Find another zone which has @min_bytes of readable data. Search in zones
1612 * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf.
1614 * Either returns NULL or returns a zone pointer. When the zone has write
1615 * pointer, hold the mutex for the zone.
1617 static struct fio_zone_info *
1618 zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes,
1619 struct fio_zone_info *zb, struct fio_zone_info *zl)
1621 struct fio_file *f = io_u->file;
1622 struct fio_zone_info *z1, *z2;
1623 const struct fio_zone_info *const zf = zbd_get_zone(f, f->min_zone);
1626 * Skip to the next non-empty zone in case of sequential I/O and to
1627 * the nearest non-empty zone in case of random I/O.
1629 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1630 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1632 zone_lock(td, f, z1);
1633 if (z1->start + min_bytes <= z1->wp)
1637 } else if (!td_random(td)) {
1641 if (td_random(td) && z2 >= zf &&
1642 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1644 zone_lock(td, f, z2);
1645 if (z2->start + min_bytes <= z2->wp)
1653 "%s: no zone has %"PRIu64" bytes of readable data\n",
1654 f->file_name, min_bytes);
1660 * zbd_end_zone_io - update zone status at command completion
1662 * @z: zone info pointer
1664 * If the write command made the zone full, remove it from the write target
1667 * The caller must hold z->mutex.
1669 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1670 struct fio_zone_info *z)
1672 const struct fio_file *f = io_u->file;
1674 if (io_u->ddir == DDIR_WRITE &&
1675 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1676 pthread_mutex_lock(&f->zbd_info->mutex);
1677 zbd_write_zone_put(td, f, z);
1678 pthread_mutex_unlock(&f->zbd_info->mutex);
1683 * zbd_queue_io - update the write pointer of a sequential zone
1685 * @success: Whether or not the I/O unit has been queued successfully
1686 * @q: queueing status (busy, completed or queued).
1688 * For write and trim operations, update the write pointer of the I/O unit
1691 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1694 const struct fio_file *f = io_u->file;
1695 struct zoned_block_device_info *zbd_info = f->zbd_info;
1696 struct fio_zone_info *z;
1701 z = zbd_offset_to_zone(f, io_u->offset);
1708 "%s: queued I/O (%lld, %llu) for zone %u\n",
1709 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1711 switch (io_u->ddir) {
1713 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1714 zbd_zone_capacity_end(z));
1717 * z->wp > zone_end means that one or more I/O errors
1720 if (accounting_vdb(td, f) && z->wp <= zone_end) {
1721 pthread_mutex_lock(&zbd_info->mutex);
1722 zbd_info->wp_valid_data_bytes += zone_end - z->wp;
1723 pthread_mutex_unlock(&zbd_info->mutex);
1731 if (q == FIO_Q_COMPLETED && !io_u->error)
1732 zbd_end_zone_io(td, io_u, z);
1735 if (!success || q != FIO_Q_QUEUED) {
1736 /* BUSY or COMPLETED: unlock the zone */
1738 io_u->zbd_put_io = NULL;
1743 * zbd_put_io - Unlock an I/O unit target zone lock
1746 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1748 const struct fio_file *f = io_u->file;
1749 struct fio_zone_info *z;
1751 assert(f->zbd_info);
1753 z = zbd_offset_to_zone(f, io_u->offset);
1757 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1758 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1760 zbd_end_zone_io(td, io_u, z);
1766 * Windows and MacOS do not define this.
1769 #define EREMOTEIO 121 /* POSIX value */
1772 bool zbd_unaligned_write(int error_code)
1774 switch (error_code) {
1783 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1784 * @td: FIO thread data.
1785 * @io_u: FIO I/O unit.
1787 * For sequential workloads, change the file offset to skip zoneskip bytes when
1788 * no more IO can be performed in the current zone.
1789 * - For read workloads, zoneskip is applied when the io has reached the end of
1790 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1791 * - For write workloads, zoneskip is applied when the zone is full.
1792 * This applies only to read and write operations.
1794 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1796 struct fio_file *f = io_u->file;
1797 enum fio_ddir ddir = io_u->ddir;
1798 struct fio_zone_info *z;
1800 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1801 assert(td->o.zone_size);
1802 assert(f->zbd_info);
1804 z = zbd_offset_to_zone(f, f->last_pos[ddir]);
1807 * When the zone capacity is smaller than the zone size and the I/O is
1808 * sequential write, skip to zone end if the latest position is at the
1809 * zone capacity limit.
1811 if (z->capacity < f->zbd_info->zone_size &&
1812 !td_random(td) && ddir == DDIR_WRITE &&
1813 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1815 "%s: Jump from zone capacity limit to zone end:"
1816 " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n",
1817 f->file_name, f->last_pos[ddir],
1818 zbd_zone_end(z), zbd_zone_idx(f, z), z->capacity);
1819 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1820 f->last_pos[ddir] = zbd_zone_end(z);
1824 * zone_skip is valid only for sequential workloads.
1826 if (td_random(td) || !td->o.zone_skip)
1830 * It is time to switch to a new zone if:
1831 * - zone_bytes == zone_size bytes have already been accessed
1832 * - The last position reached the end of the current zone.
1833 * - For reads with td->o.read_beyond_wp == false, the last position
1834 * reached the zone write pointer.
1836 if (td->zone_bytes >= td->o.zone_size ||
1837 f->last_pos[ddir] >= zbd_zone_end(z) ||
1838 (ddir == DDIR_READ &&
1839 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1844 f->file_offset += td->o.zone_size + td->o.zone_skip;
1847 * Wrap from the beginning, if we exceed the file size
1849 if (f->file_offset >= f->real_file_size)
1850 f->file_offset = get_start_offset(td, f);
1852 f->last_pos[ddir] = f->file_offset;
1853 td->io_skip_bytes += td->o.zone_skip;
1858 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1860 * @td: FIO thread data.
1861 * @io_u: FIO I/O unit.
1862 * @ddir: I/O direction before adjustment.
1864 * Return adjusted I/O direction.
1866 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1870 * In case read direction is chosen for the first random I/O, fio with
1871 * zonemode=zbd stops because no data can be read from zoned block
1872 * devices with all empty zones. Overwrite the first I/O direction as
1873 * write to make sure data to read exists.
1875 assert(io_u->file->zbd_info);
1876 if (ddir != DDIR_READ || !td_rw(td))
1879 if (io_u->file->last_start[DDIR_WRITE] != -1ULL || td->o.read_beyond_wp)
1886 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1887 * @td: FIO thread data.
1888 * @io_u: FIO I/O unit.
1890 * Locking strategy: returns with z->mutex locked if and only if z refers
1891 * to a sequential zone and if io_u_accept is returned. z is the zone that
1892 * corresponds to io_u->offset at the end of this function.
1894 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1896 struct fio_file *f = io_u->file;
1897 struct zoned_block_device_info *zbdi = f->zbd_info;
1898 struct fio_zone_info *zb, *zl, *orig_zb;
1899 uint32_t orig_len = io_u->buflen;
1900 uint64_t min_bs = td->o.min_bs[io_u->ddir];
1906 assert(is_valid_offset(f, io_u->offset));
1907 assert(io_u->buflen);
1909 zb = zbd_offset_to_zone(f, io_u->offset);
1913 /* Accept non-write I/Os for conventional zones. */
1914 if (io_u->ddir != DDIR_WRITE)
1918 * Make sure that writes to conventional zones
1919 * don't cross over to any sequential zones.
1921 if (!(zb + 1)->has_wp ||
1922 io_u->offset + io_u->buflen <= (zb + 1)->start)
1925 if (io_u->offset + min_bs > (zb + 1)->start) {
1927 "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n",
1928 f->file_name, io_u->offset,
1929 min_bs, (zb + 1)->start);
1931 zb->start + (zb + 1)->start - io_u->offset;
1932 new_len = min(io_u->buflen,
1933 (zb + 1)->start - io_u->offset);
1935 new_len = (zb + 1)->start - io_u->offset;
1938 io_u->buflen = new_len / min_bs * min_bs;
1944 * Accept the I/O offset for reads if reading beyond the write pointer
1947 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1948 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1951 zone_lock(td, f, zb);
1953 switch (io_u->ddir) {
1955 if (td->runstate == TD_VERIFYING && td_write(td))
1959 * Check that there is enough written data in the zone to do an
1960 * I/O of at least min_bs B. If there isn't, find a new zone for
1963 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1964 zb->wp - zb->start : 0;
1965 if (range < min_bs ||
1966 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1968 zl = zbd_get_zone(f, f->max_zone);
1969 zb = zbd_find_zone(td, io_u, min_bs, zb, zl);
1972 "%s: zbd_find_zone(%lld, %llu) failed\n",
1973 f->file_name, io_u->offset,
1978 * zbd_find_zone() returned a zone with a range of at
1981 range = zb->wp - zb->start;
1982 assert(range >= min_bs);
1985 io_u->offset = zb->start;
1989 * Make sure the I/O is within the zone valid data range while
1990 * maximizing the I/O size and preserving randomness.
1992 if (range <= io_u->buflen)
1993 io_u->offset = zb->start;
1994 else if (td_random(td))
1995 io_u->offset = zb->start +
1996 ((io_u->offset - orig_zb->start) %
1997 (range - io_u->buflen)) / min_bs * min_bs;
2000 * When zbd_find_zone() returns a conventional zone,
2001 * we can simply accept the new i/o offset here.
2007 * Make sure the I/O does not cross over the zone wp position.
2009 new_len = min((unsigned long long)io_u->buflen,
2010 (unsigned long long)(zb->wp - io_u->offset));
2011 new_len = new_len / min_bs * min_bs;
2012 if (new_len < io_u->buflen) {
2013 io_u->buflen = new_len;
2014 dprint(FD_IO, "Changed length from %u into %llu\n",
2015 orig_len, io_u->buflen);
2018 assert(zb->start <= io_u->offset);
2019 assert(io_u->offset + io_u->buflen <= zb->wp);
2024 if (io_u->buflen > zbdi->zone_size) {
2025 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
2027 "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n",
2028 f->file_name, io_u->buflen, zbdi->zone_size);
2033 if (zbd_zone_remainder(zb) > 0 &&
2034 zbd_zone_remainder(zb) < min_bs) {
2035 pthread_mutex_lock(&f->zbd_info->mutex);
2036 zbd_write_zone_put(td, f, zb);
2037 pthread_mutex_unlock(&f->zbd_info->mutex);
2039 "%s: finish zone %d\n",
2040 f->file_name, zbd_zone_idx(f, zb));
2042 zbd_finish_zone(td, f, zb);
2043 if (zbd_zone_idx(f, zb) + 1 >= f->max_zone) {
2049 /* Find the next write pointer zone */
2052 if (zbd_zone_idx(f, zb) >= f->max_zone)
2053 zb = zbd_get_zone(f, f->min_zone);
2054 } while (!zb->has_wp);
2056 zone_lock(td, f, zb);
2059 if (!zbd_write_zone_get(td, f, zb)) {
2061 zb = zbd_convert_to_write_zone(td, io_u);
2063 dprint(FD_IO, "%s: can't convert to write target zone",
2069 if (zbd_zone_remainder(zb) > 0 &&
2070 zbd_zone_remainder(zb) < min_bs)
2073 /* Check whether the zone reset threshold has been exceeded */
2074 if (td->o.zrf.u.f) {
2075 if (zbdi->wp_valid_data_bytes >=
2076 f->io_size * td->o.zrt.u.f &&
2077 zbd_dec_and_reset_write_cnt(td, f))
2081 /* Reset the zone pointer if necessary */
2082 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
2083 if (td->o.verify != VERIFY_NONE) {
2085 * Unset io-u->file to tell get_next_verify()
2086 * that this IO is not requeue.
2089 if (!get_next_verify(td, io_u)) {
2097 * Since previous write requests may have been submitted
2098 * asynchronously and since we will submit the zone
2099 * reset synchronously, wait until previously submitted
2100 * write requests have completed before issuing a
2105 if (__zbd_reset_zone(td, f, zb) < 0)
2108 if (zb->capacity < min_bs) {
2109 td_verror(td, EINVAL, "ZCAP is less min_bs");
2110 log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n",
2111 zb->capacity, min_bs);
2116 /* Make writes occur at the write pointer */
2117 assert(!zbd_zone_full(f, zb, min_bs));
2118 io_u->offset = zb->wp;
2119 if (!is_valid_offset(f, io_u->offset)) {
2120 td_verror(td, EINVAL, "invalid WP value");
2121 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
2122 f->file_name, io_u->offset);
2127 * Make sure that the buflen is a multiple of the minimal
2128 * block size. Give up if shrinking would make the request too
2131 new_len = min((unsigned long long)io_u->buflen,
2132 zbd_zone_capacity_end(zb) - io_u->offset);
2133 new_len = new_len / min_bs * min_bs;
2134 if (new_len == io_u->buflen)
2136 if (new_len >= min_bs) {
2137 io_u->buflen = new_len;
2138 dprint(FD_IO, "Changed length from %u into %llu\n",
2139 orig_len, io_u->buflen);
2143 td_verror(td, EIO, "zone remainder too small");
2144 log_err("zone remainder %lld smaller than min block size %"PRIu64"\n",
2145 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
2150 /* Check random trim targets a non-empty zone */
2151 if (!td_random(td) || zb->wp > zb->start)
2154 /* Find out a non-empty zone to trim */
2156 zl = zbd_get_zone(f, f->max_zone);
2157 zb = zbd_find_zone(td, io_u, 1, zb, zl);
2159 io_u->offset = zb->start;
2160 dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n",
2161 f->file_name, io_u->offset);
2170 case DDIR_SYNC_FILE_RANGE:
2181 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
2182 assert(!io_u->zbd_queue_io);
2183 assert(!io_u->zbd_put_io);
2185 io_u->zbd_queue_io = zbd_queue_io;
2186 io_u->zbd_put_io = zbd_put_io;
2189 * Since we return with the zone lock still held,
2190 * add an annotation to let Coverity know that it
2193 /* coverity[missing_unlock] */
2198 if (zb && zb->has_wp)
2204 /* Return a string with ZBD statistics */
2205 char *zbd_write_status(const struct thread_stat *ts)
2209 if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0)
2215 * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim
2217 * @td: FIO thread data.
2218 * @io_u: FIO I/O unit.
2220 * It is assumed that z->mutex is already locked.
2221 * Return io_u_completed when reset zone succeeds. Return 0 when the target zone
2222 * does not have write pointer. On error, return negative errno.
2224 int zbd_do_io_u_trim(struct thread_data *td, struct io_u *io_u)
2226 struct fio_file *f = io_u->file;
2227 struct fio_zone_info *z;
2230 z = zbd_offset_to_zone(f, io_u->offset);
2234 if (io_u->offset != z->start) {
2235 log_err("Trim offset not at zone start (%lld)\n",
2240 ret = zbd_reset_zone((struct thread_data *)td, f, z);
2244 return io_u_completed;
2247 void zbd_log_err(const struct thread_data *td, const struct io_u *io_u)
2249 const struct fio_file *f = io_u->file;
2251 if (td->o.zone_mode != ZONE_MODE_ZBD)
2254 if (io_u->error == EOVERFLOW)
2255 log_err("%s: Exceeded max_active_zones limit. Check conditions of zones out of I/O ranges.\n",