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_reset_zones - Reset a range of zones.
339 * @td: fio thread data.
340 * @f: fio file for which to reset zones
341 * @zb: first zone to reset.
342 * @ze: first zone not to reset.
344 * Returns 0 upon success and 1 upon failure.
346 static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
347 struct fio_zone_info *const zb,
348 struct fio_zone_info *const ze)
350 struct fio_zone_info *z;
351 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
356 dprint(FD_ZBD, "%s: examining zones %u .. %u\n",
357 f->file_name, zbd_zone_idx(f, zb), zbd_zone_idx(f, ze));
359 for (z = zb; z < ze; z++) {
364 pthread_mutex_lock(&f->zbd_info->mutex);
365 zbd_close_zone(td, f, z);
366 pthread_mutex_unlock(&f->zbd_info->mutex);
368 if (z->wp != z->start) {
369 dprint(FD_ZBD, "%s: resetting zone %u\n",
370 f->file_name, zbd_zone_idx(f, z));
371 if (zbd_reset_zone(td, f, z) < 0)
382 * zbd_get_max_open_zones - Get the maximum number of open zones
383 * @td: FIO thread data
384 * @f: FIO file for which to get max open zones
385 * @max_open_zones: Upon success, result will be stored here.
387 * A @max_open_zones value set to zero means no limit.
389 * Returns 0 upon success and a negative error code upon failure.
391 static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f,
392 unsigned int *max_open_zones)
396 if (td->io_ops && td->io_ops->get_max_open_zones)
397 ret = td->io_ops->get_max_open_zones(td, f, max_open_zones);
399 ret = blkzoned_get_max_open_zones(td, f, max_open_zones);
401 td_verror(td, errno, "get max open zones failed");
402 log_err("%s: get max open zones failed (%d).\n",
403 f->file_name, errno);
410 * zbd_open_zone - Add a zone to the array of open zones.
411 * @td: fio thread data.
412 * @f: fio file that has the open zones to add.
413 * @zone_idx: Index of the zone to add.
415 * Open a ZBD zone if it is not already open. Returns true if either the zone
416 * was already open or if the zone was successfully added to the array of open
417 * zones without exceeding the maximum number of open zones. Returns false if
418 * the zone was not already open and opening the zone would cause the zone limit
421 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
422 struct fio_zone_info *z)
424 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
425 struct zoned_block_device_info *zbdi = f->zbd_info;
426 uint32_t zone_idx = zbd_zone_idx(f, z);
429 if (z->cond == ZBD_ZONE_COND_OFFLINE)
433 * Skip full zones with data verification enabled because resetting a
434 * zone causes data loss and hence causes verification to fail.
436 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
440 * zbdi->max_open_zones == 0 means that there is no limit on the maximum
441 * number of open zones. In this case, do no track open zones in
442 * zbdi->open_zones array.
444 if (!zbdi->max_open_zones)
447 pthread_mutex_lock(&zbdi->mutex);
451 * If the zone is going to be completely filled by writes
452 * already in-flight, handle it as a full zone instead of an
455 if (!zbd_zone_remainder(z))
461 /* Zero means no limit */
462 if (td->o.job_max_open_zones > 0 &&
463 td->num_open_zones >= td->o.job_max_open_zones)
465 if (zbdi->num_open_zones >= zbdi->max_open_zones)
468 dprint(FD_ZBD, "%s: opening zone %u\n",
469 f->file_name, zone_idx);
471 zbdi->open_zones[zbdi->num_open_zones++] = zone_idx;
472 td->num_open_zones++;
477 pthread_mutex_unlock(&zbdi->mutex);
481 /* Verify whether direct I/O is used for all host-managed zoned block drives. */
482 static bool zbd_using_direct_io(void)
484 struct thread_data *td;
489 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
491 for_each_file(td, f, j) {
492 if (f->zbd_info && f->filetype == FIO_TYPE_BLOCK &&
493 f->zbd_info->model == ZBD_HOST_MANAGED)
501 /* Whether or not the I/O range for f includes one or more sequential zones */
502 static bool zbd_is_seq_job(struct fio_file *f)
504 uint32_t zone_idx, zone_idx_b, zone_idx_e;
511 zone_idx_b = zbd_offset_to_zone_idx(f, f->file_offset);
513 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size - 1);
514 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
515 if (zbd_get_zone(f, zone_idx)->has_wp)
522 * Verify whether the file offset and size parameters are aligned with zone
523 * boundaries. If the file offset is not aligned, align it down to the start of
524 * the zone containing the start offset and align up the file io_size parameter.
526 static bool zbd_zone_align_file_sizes(struct thread_data *td,
529 const struct fio_zone_info *z;
530 uint64_t new_offset, new_end;
534 if (f->file_offset >= f->real_file_size)
536 if (!zbd_is_seq_job(f))
539 if (!td->o.zone_size) {
540 td->o.zone_size = f->zbd_info->zone_size;
541 if (!td->o.zone_size) {
542 log_err("%s: invalid 0 zone size\n",
546 } else if (td->o.zone_size != f->zbd_info->zone_size) {
547 log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n",
548 f->file_name, td->o.zone_size,
549 f->zbd_info->zone_size);
553 if (td->o.zone_skip % td->o.zone_size) {
554 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
555 f->file_name, td->o.zone_skip,
560 z = zbd_offset_to_zone(f, f->file_offset);
561 if ((f->file_offset != z->start) &&
562 (td->o.td_ddir != TD_DDIR_READ)) {
563 new_offset = zbd_zone_end(z);
564 if (new_offset >= f->file_offset + f->io_size) {
565 log_info("%s: io_size must be at least one zone\n",
569 log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n",
570 f->file_name, f->file_offset,
572 f->io_size -= (new_offset - f->file_offset);
573 f->file_offset = new_offset;
576 z = zbd_offset_to_zone(f, f->file_offset + f->io_size);
578 if ((td->o.td_ddir != TD_DDIR_READ) &&
579 (f->file_offset + f->io_size != new_end)) {
580 if (new_end <= f->file_offset) {
581 log_info("%s: io_size must be at least one zone\n",
585 log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n",
586 f->file_name, f->io_size,
587 new_end - f->file_offset);
588 f->io_size = new_end - f->file_offset;
595 * Verify whether offset and size parameters are aligned with zone boundaries.
597 static bool zbd_verify_sizes(void)
599 struct thread_data *td;
604 for_each_file(td, f, j) {
605 if (!zbd_zone_align_file_sizes(td, f))
613 static bool zbd_verify_bs(void)
615 struct thread_data *td;
621 (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] ||
622 td->o.bssplit_nr[DDIR_TRIM])) {
623 log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n");
626 for_each_file(td, f, j) {
632 zone_size = f->zbd_info->zone_size;
633 if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) {
634 log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n",
635 f->file_name, td->o.bs[DDIR_TRIM],
639 for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) {
640 if (td->o.verify != VERIFY_NONE &&
641 zone_size % td->o.bs[k] != 0) {
642 log_info("%s: block size %llu is not a divisor of the zone size %"PRIu64"\n",
643 f->file_name, td->o.bs[k],
653 static int ilog2(uint64_t i)
665 * Initialize f->zbd_info for devices that are not zoned block devices. This
666 * allows to execute a ZBD workload against a non-ZBD device.
668 static int init_zone_info(struct thread_data *td, struct fio_file *f)
671 struct fio_zone_info *p;
672 uint64_t zone_size = td->o.zone_size;
673 uint64_t zone_capacity = td->o.zone_capacity;
674 struct zoned_block_device_info *zbd_info = NULL;
677 if (zone_size == 0) {
678 log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n",
683 if (zone_size < 512) {
684 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
689 if (zone_capacity == 0)
690 zone_capacity = zone_size;
692 if (zone_capacity > zone_size) {
693 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
694 f->file_name, td->o.zone_capacity, td->o.zone_size);
698 if (f->real_file_size < zone_size) {
699 log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n",
700 f->file_name, f->real_file_size, zone_size);
704 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
705 zbd_info = scalloc(1, sizeof(*zbd_info) +
706 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
710 mutex_init_pshared(&zbd_info->mutex);
711 zbd_info->refcount = 1;
712 p = &zbd_info->zone_info[0];
713 for (i = 0; i < nr_zones; i++, p++) {
714 mutex_init_pshared_with_type(&p->mutex,
715 PTHREAD_MUTEX_RECURSIVE);
716 p->start = i * zone_size;
718 p->type = ZBD_ZONE_TYPE_SWR;
719 p->cond = ZBD_ZONE_COND_EMPTY;
720 p->capacity = zone_capacity;
724 p->start = nr_zones * zone_size;
726 f->zbd_info = zbd_info;
727 f->zbd_info->zone_size = zone_size;
728 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
729 ilog2(zone_size) : 0;
730 f->zbd_info->nr_zones = nr_zones;
735 * Maximum number of zones to report in one operation.
737 #define ZBD_REPORT_MAX_ZONES 8192U
740 * Parse the device zone report and store it in f->zbd_info. Must be called
741 * only for devices that are zoned, namely those with a model != ZBD_NONE.
743 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
746 struct zbd_zone *zones, *z;
747 struct fio_zone_info *p;
748 uint64_t zone_size, offset;
749 struct zoned_block_device_info *zbd_info = NULL;
750 int i, j, ret = -ENOMEM;
752 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
756 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
759 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
764 zone_size = zones[0].len;
765 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
767 if (td->o.zone_size == 0) {
768 td->o.zone_size = zone_size;
769 } else if (td->o.zone_size != zone_size) {
770 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n",
771 f->file_name, td->o.zone_size, zone_size);
776 dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n",
777 f->file_name, nr_zones, zone_size / 1024);
779 zbd_info = scalloc(1, sizeof(*zbd_info) +
780 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
783 mutex_init_pshared(&zbd_info->mutex);
784 zbd_info->refcount = 1;
785 p = &zbd_info->zone_info[0];
786 for (offset = 0, j = 0; j < nr_zones;) {
788 for (i = 0; i < nrz; i++, j++, z++, p++) {
789 mutex_init_pshared_with_type(&p->mutex,
790 PTHREAD_MUTEX_RECURSIVE);
792 p->capacity = z->capacity;
795 case ZBD_ZONE_COND_NOT_WP:
796 case ZBD_ZONE_COND_FULL:
797 p->wp = p->start + p->capacity;
800 assert(z->start <= z->wp);
801 assert(z->wp <= z->start + zone_size);
807 case ZBD_ZONE_TYPE_SWR:
816 if (j > 0 && p->start != p[-1].start + zone_size) {
817 log_info("%s: invalid zone data\n",
824 offset = z->start + z->len;
828 nrz = zbd_report_zones(td, f, offset, zones,
829 min((uint32_t)(nr_zones - j),
830 ZBD_REPORT_MAX_ZONES));
833 log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n",
834 offset, f->file_name, -ret);
840 zbd_info->zone_info[nr_zones].start = offset;
842 f->zbd_info = zbd_info;
843 f->zbd_info->zone_size = zone_size;
844 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
845 ilog2(zone_size) : 0;
846 f->zbd_info->nr_zones = nr_zones;
856 static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f)
858 struct zoned_block_device_info *zbd = f->zbd_info;
859 unsigned int max_open_zones;
862 if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) {
863 /* Only host-managed devices have a max open limit */
864 zbd->max_open_zones = td->o.max_open_zones;
868 /* If host-managed, get the max open limit */
869 ret = zbd_get_max_open_zones(td, f, &max_open_zones);
873 if (!max_open_zones) {
874 /* No device limit */
875 zbd->max_open_zones = td->o.max_open_zones;
876 } else if (!td->o.max_open_zones) {
877 /* No user limit. Set limit to device limit */
878 zbd->max_open_zones = max_open_zones;
879 } else if (td->o.max_open_zones <= max_open_zones) {
880 /* Both user limit and dev limit. User limit not too large */
881 zbd->max_open_zones = td->o.max_open_zones;
883 /* Both user limit and dev limit. User limit too large */
884 td_verror(td, EINVAL,
885 "Specified --max_open_zones is too large");
886 log_err("Specified --max_open_zones (%d) is larger than max (%u)\n",
887 td->o.max_open_zones, max_open_zones);
892 /* Ensure that the limit is not larger than FIO's internal limit */
893 if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) {
894 td_verror(td, EINVAL, "'max_open_zones' value is too large");
895 log_err("'max_open_zones' value is larger than %u\n",
900 dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n",
901 f->file_name, zbd->max_open_zones);
907 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
909 * Returns 0 upon success and a negative error code upon failure.
911 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
913 enum zbd_zoned_model zbd_model;
916 assert(td->o.zone_mode == ZONE_MODE_ZBD);
918 ret = zbd_get_zoned_model(td, f, &zbd_model);
924 case ZBD_HOST_MANAGED:
925 ret = parse_zone_info(td, f);
930 ret = init_zone_info(td, f);
935 td_verror(td, EINVAL, "Unsupported zoned model");
936 log_err("Unsupported zoned model\n");
941 f->zbd_info->model = zbd_model;
943 ret = zbd_set_max_open_zones(td, f);
945 zbd_free_zone_info(f);
952 void zbd_free_zone_info(struct fio_file *f)
958 pthread_mutex_lock(&f->zbd_info->mutex);
959 refcount = --f->zbd_info->refcount;
960 pthread_mutex_unlock(&f->zbd_info->mutex);
962 assert((int32_t)refcount >= 0);
969 * Initialize f->zbd_info.
971 * Returns 0 upon success and a negative error code upon failure.
973 * Note: this function can only work correctly if it is called before the first
976 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
978 struct thread_data *td2;
982 for_each_td(td2, i) {
983 for_each_file(td2, f2, j) {
984 if (td2 == td && f2 == file)
987 strcmp(f2->file_name, file->file_name) != 0)
989 file->zbd_info = f2->zbd_info;
990 file->zbd_info->refcount++;
995 ret = zbd_create_zone_info(td, file);
997 td_verror(td, -ret, "zbd_create_zone_info() failed");
1002 int zbd_init_files(struct thread_data *td)
1007 for_each_file(td, f, i) {
1008 if (zbd_init_zone_info(td, f))
1015 void zbd_recalc_options_with_zone_granularity(struct thread_data *td)
1020 for_each_file(td, f, i) {
1021 struct zoned_block_device_info *zbd = f->zbd_info;
1024 /* zonemode=strided doesn't get per-file zone size. */
1025 zone_size = zbd ? zbd->zone_size : td->o.zone_size;
1029 if (td->o.size_nz > 0)
1030 td->o.size = td->o.size_nz * zone_size;
1031 if (td->o.io_size_nz > 0)
1032 td->o.io_size = td->o.io_size_nz * zone_size;
1033 if (td->o.start_offset_nz > 0)
1034 td->o.start_offset = td->o.start_offset_nz * zone_size;
1035 if (td->o.offset_increment_nz > 0)
1036 td->o.offset_increment =
1037 td->o.offset_increment_nz * zone_size;
1038 if (td->o.zone_skip_nz > 0)
1039 td->o.zone_skip = td->o.zone_skip_nz * zone_size;
1043 int zbd_setup_files(struct thread_data *td)
1048 if (!zbd_using_direct_io()) {
1049 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
1053 if (!zbd_verify_sizes())
1056 if (!zbd_verify_bs())
1059 for_each_file(td, f, i) {
1060 struct zoned_block_device_info *zbd = f->zbd_info;
1061 struct fio_zone_info *z;
1066 f->min_zone = zbd_offset_to_zone_idx(f, f->file_offset);
1068 zbd_offset_to_zone_idx(f, f->file_offset + f->io_size);
1071 * When all zones in the I/O range are conventional, io_size
1072 * can be smaller than zone size, making min_zone the same
1073 * as max_zone. This is why the assert below needs to be made
1076 if (zbd_is_seq_job(f))
1077 assert(f->min_zone < f->max_zone);
1079 if (td->o.max_open_zones > 0 &&
1080 zbd->max_open_zones != td->o.max_open_zones) {
1081 log_err("Different 'max_open_zones' values\n");
1086 * The per job max open zones limit cannot be used without a
1087 * global max open zones limit. (As the tracking of open zones
1088 * is disabled when there is no global max open zones limit.)
1090 if (td->o.job_max_open_zones && !zbd->max_open_zones) {
1091 log_err("'job_max_open_zones' cannot be used without a global open zones limit\n");
1096 * zbd->max_open_zones is the global limit shared for all jobs
1097 * that target the same zoned block device. Force sync the per
1098 * thread global limit with the actual global limit. (The real
1099 * per thread/job limit is stored in td->o.job_max_open_zones).
1101 td->o.max_open_zones = zbd->max_open_zones;
1103 for (zi = f->min_zone; zi < f->max_zone; zi++) {
1104 z = &zbd->zone_info[zi];
1105 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
1106 z->cond != ZBD_ZONE_COND_EXP_OPEN)
1108 if (zbd_open_zone(td, f, z))
1111 * If the number of open zones exceeds specified limits,
1112 * reset all extra open zones.
1114 if (zbd_reset_zone(td, f, z) < 0) {
1115 log_err("Failed to reest zone %d\n", zi);
1125 * Reset zbd_info.write_cnt, the counter that counts down towards the next
1128 static void _zbd_reset_write_cnt(const struct thread_data *td,
1129 const struct fio_file *f)
1131 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
1133 f->zbd_info->write_cnt = td->o.zrf.u.f ?
1134 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
1137 static void zbd_reset_write_cnt(const struct thread_data *td,
1138 const struct fio_file *f)
1140 pthread_mutex_lock(&f->zbd_info->mutex);
1141 _zbd_reset_write_cnt(td, f);
1142 pthread_mutex_unlock(&f->zbd_info->mutex);
1145 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
1146 const struct fio_file *f)
1148 uint32_t write_cnt = 0;
1150 pthread_mutex_lock(&f->zbd_info->mutex);
1151 assert(f->zbd_info->write_cnt);
1152 if (f->zbd_info->write_cnt)
1153 write_cnt = --f->zbd_info->write_cnt;
1155 _zbd_reset_write_cnt(td, f);
1156 pthread_mutex_unlock(&f->zbd_info->mutex);
1158 return write_cnt == 0;
1166 /* Calculate the number of sectors with data (swd) and perform action 'a' */
1167 static uint64_t zbd_process_swd(struct thread_data *td,
1168 const struct fio_file *f, enum swd_action a)
1170 struct fio_zone_info *zb, *ze, *z;
1172 uint64_t wp_swd = 0;
1174 zb = zbd_get_zone(f, f->min_zone);
1175 ze = zbd_get_zone(f, f->max_zone);
1176 for (z = zb; z < ze; z++) {
1178 zone_lock(td, f, z);
1179 wp_swd += z->wp - z->start;
1181 swd += z->wp - z->start;
1184 pthread_mutex_lock(&f->zbd_info->mutex);
1187 assert(f->zbd_info->sectors_with_data == swd);
1188 assert(f->zbd_info->wp_sectors_with_data == wp_swd);
1191 f->zbd_info->sectors_with_data = swd;
1192 f->zbd_info->wp_sectors_with_data = wp_swd;
1195 pthread_mutex_unlock(&f->zbd_info->mutex);
1197 for (z = zb; z < ze; z++)
1205 * The swd check is useful for debugging but takes too much time to leave
1206 * it enabled all the time. Hence it is disabled by default.
1208 static const bool enable_check_swd = false;
1210 /* Check whether the values of zbd_info.*sectors_with_data are correct. */
1211 static void zbd_check_swd(struct thread_data *td, const struct fio_file *f)
1213 if (!enable_check_swd)
1216 zbd_process_swd(td, f, CHECK_SWD);
1219 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
1221 struct fio_zone_info *zb, *ze;
1224 if (!f->zbd_info || !td_write(td))
1227 zb = zbd_get_zone(f, f->min_zone);
1228 ze = zbd_get_zone(f, f->max_zone);
1229 swd = zbd_process_swd(td, f, SET_SWD);
1231 dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n",
1232 __func__, f->file_name, swd);
1235 * If data verification is enabled reset the affected zones before
1236 * writing any data to avoid that a zone reset has to be issued while
1237 * writing data, which causes data loss.
1239 if (td->o.verify != VERIFY_NONE && td->runstate != TD_VERIFYING)
1240 zbd_reset_zones(td, f, zb, ze);
1241 zbd_reset_write_cnt(td, f);
1244 /* Return random zone index for one of the open zones. */
1245 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1246 const struct io_u *io_u)
1248 return (io_u->offset - f->file_offset) *
1249 f->zbd_info->num_open_zones / f->io_size;
1252 static bool any_io_in_flight(void)
1254 struct thread_data *td;
1257 for_each_td(td, i) {
1258 if (td->io_u_in_flight)
1266 * Modify the offset of an I/O unit that does not refer to an open zone such
1267 * that it refers to an open zone. Close an open zone and open a new zone if
1268 * necessary. The open zone is searched across sequential zones.
1269 * This algorithm can only work correctly if all write pointers are
1270 * a multiple of the fio block size. The caller must neither hold z->mutex
1271 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1273 static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
1276 const uint64_t min_bs = td->o.min_bs[io_u->ddir];
1277 struct fio_file *f = io_u->file;
1278 struct zoned_block_device_info *zbdi = f->zbd_info;
1279 struct fio_zone_info *z;
1280 unsigned int open_zone_idx = -1;
1281 uint32_t zone_idx, new_zone_idx;
1283 bool wait_zone_close;
1285 bool should_retry = true;
1287 assert(is_valid_offset(f, io_u->offset));
1289 if (zbdi->max_open_zones || td->o.job_max_open_zones) {
1291 * This statement accesses zbdi->open_zones[] on purpose
1294 zone_idx = zbdi->open_zones[pick_random_zone_idx(f, io_u)];
1296 zone_idx = zbd_offset_to_zone_idx(f, io_u->offset);
1298 if (zone_idx < f->min_zone)
1299 zone_idx = f->min_zone;
1300 else if (zone_idx >= f->max_zone)
1301 zone_idx = f->max_zone - 1;
1304 "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1305 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1308 * Since z->mutex is the outer lock and zbdi->mutex the inner
1309 * lock it can happen that the state of the zone with index zone_idx
1310 * has changed after 'z' has been assigned and before zbdi->mutex
1311 * has been obtained. Hence the loop.
1316 z = zbd_get_zone(f, zone_idx);
1318 zone_lock(td, f, z);
1320 pthread_mutex_lock(&zbdi->mutex);
1323 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1324 zbdi->max_open_zones == 0 &&
1325 td->o.job_max_open_zones == 0)
1327 if (zbdi->num_open_zones == 0) {
1328 dprint(FD_ZBD, "%s(%s): no zones are open\n",
1329 __func__, f->file_name);
1330 goto open_other_zone;
1335 * List of opened zones is per-device, shared across all
1336 * threads. Start with quasi-random candidate zone. Ignore
1337 * zones which don't belong to thread's offset/size area.
1339 open_zone_idx = pick_random_zone_idx(f, io_u);
1340 assert(!open_zone_idx ||
1341 open_zone_idx < zbdi->num_open_zones);
1342 tmp_idx = open_zone_idx;
1344 for (i = 0; i < zbdi->num_open_zones; i++) {
1347 if (tmp_idx >= zbdi->num_open_zones)
1349 tmpz = zbdi->open_zones[tmp_idx];
1350 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1351 open_zone_idx = tmp_idx;
1352 goto found_candidate_zone;
1358 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1359 __func__, f->file_name);
1361 pthread_mutex_unlock(&zbdi->mutex);
1368 found_candidate_zone:
1369 new_zone_idx = zbdi->open_zones[open_zone_idx];
1370 if (new_zone_idx == zone_idx)
1372 zone_idx = new_zone_idx;
1374 pthread_mutex_unlock(&zbdi->mutex);
1380 /* Both z->mutex and zbdi->mutex are held. */
1383 if (zbd_zone_remainder(z) >= min_bs) {
1384 pthread_mutex_unlock(&zbdi->mutex);
1389 /* Check if number of open zones reaches one of limits. */
1391 zbdi->num_open_zones == f->max_zone - f->min_zone ||
1392 (zbdi->max_open_zones &&
1393 zbdi->num_open_zones == zbdi->max_open_zones) ||
1394 (td->o.job_max_open_zones &&
1395 td->num_open_zones == td->o.job_max_open_zones);
1397 pthread_mutex_unlock(&zbdi->mutex);
1399 /* Only z->mutex is held. */
1402 * When number of open zones reaches to one of limits, wait for
1403 * zone close before opening a new zone.
1405 if (wait_zone_close) {
1407 "%s(%s): quiesce to allow open zones to close\n",
1408 __func__, f->file_name);
1413 /* Zone 'z' is full, so try to open a new zone. */
1414 for (i = f->io_size / zbdi->zone_size; i > 0; i--) {
1419 if (!is_valid_offset(f, z->start)) {
1421 zone_idx = f->min_zone;
1422 z = zbd_get_zone(f, zone_idx);
1424 assert(is_valid_offset(f, z->start));
1427 zone_lock(td, f, z);
1430 if (zbd_open_zone(td, f, z))
1434 /* Only z->mutex is held. */
1436 /* Check whether the write fits in any of the already opened zones. */
1437 pthread_mutex_lock(&zbdi->mutex);
1438 for (i = 0; i < zbdi->num_open_zones; i++) {
1439 zone_idx = zbdi->open_zones[i];
1440 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1442 pthread_mutex_unlock(&zbdi->mutex);
1445 z = zbd_get_zone(f, zone_idx);
1447 zone_lock(td, f, z);
1448 if (zbd_zone_remainder(z) >= min_bs)
1450 pthread_mutex_lock(&zbdi->mutex);
1454 * When any I/O is in-flight or when all I/Os in-flight get completed,
1455 * the I/Os might have closed zones then retry the steps to open a zone.
1456 * Before retry, call io_u_quiesce() to complete in-flight writes.
1458 in_flight = any_io_in_flight();
1459 if (in_flight || should_retry) {
1461 "%s(%s): wait zone close and retry open zones\n",
1462 __func__, f->file_name);
1463 pthread_mutex_unlock(&zbdi->mutex);
1466 zone_lock(td, f, z);
1467 should_retry = in_flight;
1471 pthread_mutex_unlock(&zbdi->mutex);
1475 dprint(FD_ZBD, "%s(%s): did not open another zone\n",
1476 __func__, f->file_name);
1481 dprint(FD_ZBD, "%s(%s): returning zone %d\n",
1482 __func__, f->file_name, zone_idx);
1484 io_u->offset = z->start;
1486 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1491 /* The caller must hold z->mutex. */
1492 static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td,
1494 struct fio_zone_info *z)
1496 const struct fio_file *f = io_u->file;
1497 const uint64_t min_bs = td->o.min_bs[DDIR_WRITE];
1499 if (!zbd_open_zone(td, f, z)) {
1501 z = zbd_convert_to_open_zone(td, io_u);
1505 if (z->verify_block * min_bs >= z->capacity) {
1506 log_err("%s: %d * %"PRIu64" >= %"PRIu64"\n",
1507 f->file_name, z->verify_block, min_bs, z->capacity);
1509 * If the assertion below fails during a test run, adding
1510 * "--experimental_verify=1" to the command line may help.
1515 io_u->offset = z->start + z->verify_block * min_bs;
1516 if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1517 log_err("%s: %llu + %llu >= %"PRIu64"\n",
1518 f->file_name, io_u->offset, io_u->buflen,
1519 zbd_zone_capacity_end(z));
1522 z->verify_block += io_u->buflen / min_bs;
1528 * Find another zone which has @min_bytes of readable data. Search in zones
1529 * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf.
1531 * Either returns NULL or returns a zone pointer. When the zone has write
1532 * pointer, hold the mutex for the zone.
1534 static struct fio_zone_info *
1535 zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes,
1536 struct fio_zone_info *zb, struct fio_zone_info *zl)
1538 struct fio_file *f = io_u->file;
1539 struct fio_zone_info *z1, *z2;
1540 const struct fio_zone_info *const zf = zbd_get_zone(f, f->min_zone);
1543 * Skip to the next non-empty zone in case of sequential I/O and to
1544 * the nearest non-empty zone in case of random I/O.
1546 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1547 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1549 zone_lock(td, f, z1);
1550 if (z1->start + min_bytes <= z1->wp)
1554 } else if (!td_random(td)) {
1558 if (td_random(td) && z2 >= zf &&
1559 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1561 zone_lock(td, f, z2);
1562 if (z2->start + min_bytes <= z2->wp)
1570 "%s: no zone has %"PRIu64" bytes of readable data\n",
1571 f->file_name, min_bytes);
1577 * zbd_end_zone_io - update zone status at command completion
1579 * @z: zone info pointer
1581 * If the write command made the zone full, close it.
1583 * The caller must hold z->mutex.
1585 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1586 struct fio_zone_info *z)
1588 const struct fio_file *f = io_u->file;
1590 if (io_u->ddir == DDIR_WRITE &&
1591 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1592 pthread_mutex_lock(&f->zbd_info->mutex);
1593 zbd_close_zone(td, f, z);
1594 pthread_mutex_unlock(&f->zbd_info->mutex);
1599 * zbd_queue_io - update the write pointer of a sequential zone
1601 * @success: Whether or not the I/O unit has been queued successfully
1602 * @q: queueing status (busy, completed or queued).
1604 * For write and trim operations, update the write pointer of the I/O unit
1607 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1610 const struct fio_file *f = io_u->file;
1611 struct zoned_block_device_info *zbd_info = f->zbd_info;
1612 struct fio_zone_info *z;
1617 z = zbd_offset_to_zone(f, io_u->offset);
1624 "%s: queued I/O (%lld, %llu) for zone %u\n",
1625 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1627 switch (io_u->ddir) {
1629 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1630 zbd_zone_capacity_end(z));
1633 * z->wp > zone_end means that one or more I/O errors
1636 pthread_mutex_lock(&zbd_info->mutex);
1637 if (z->wp <= zone_end) {
1638 zbd_info->sectors_with_data += zone_end - z->wp;
1639 zbd_info->wp_sectors_with_data += zone_end - z->wp;
1641 pthread_mutex_unlock(&zbd_info->mutex);
1648 if (q == FIO_Q_COMPLETED && !io_u->error)
1649 zbd_end_zone_io(td, io_u, z);
1652 if (!success || q != FIO_Q_QUEUED) {
1653 /* BUSY or COMPLETED: unlock the zone */
1655 io_u->zbd_put_io = NULL;
1660 * zbd_put_io - Unlock an I/O unit target zone lock
1663 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1665 const struct fio_file *f = io_u->file;
1666 struct zoned_block_device_info *zbd_info = f->zbd_info;
1667 struct fio_zone_info *z;
1671 z = zbd_offset_to_zone(f, io_u->offset);
1675 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1676 f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z));
1678 zbd_end_zone_io(td, io_u, z);
1681 zbd_check_swd(td, f);
1685 * Windows and MacOS do not define this.
1688 #define EREMOTEIO 121 /* POSIX value */
1691 bool zbd_unaligned_write(int error_code)
1693 switch (error_code) {
1702 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1703 * @td: FIO thread data.
1704 * @io_u: FIO I/O unit.
1706 * For sequential workloads, change the file offset to skip zoneskip bytes when
1707 * no more IO can be performed in the current zone.
1708 * - For read workloads, zoneskip is applied when the io has reached the end of
1709 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1710 * - For write workloads, zoneskip is applied when the zone is full.
1711 * This applies only to read and write operations.
1713 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1715 struct fio_file *f = io_u->file;
1716 enum fio_ddir ddir = io_u->ddir;
1717 struct fio_zone_info *z;
1719 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1720 assert(td->o.zone_size);
1721 assert(f->zbd_info);
1723 z = zbd_offset_to_zone(f, f->last_pos[ddir]);
1726 * When the zone capacity is smaller than the zone size and the I/O is
1727 * sequential write, skip to zone end if the latest position is at the
1728 * zone capacity limit.
1730 if (z->capacity < f->zbd_info->zone_size &&
1731 !td_random(td) && ddir == DDIR_WRITE &&
1732 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1734 "%s: Jump from zone capacity limit to zone end:"
1735 " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n",
1736 f->file_name, f->last_pos[ddir],
1737 zbd_zone_end(z), zbd_zone_idx(f, z), z->capacity);
1738 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1739 f->last_pos[ddir] = zbd_zone_end(z);
1743 * zone_skip is valid only for sequential workloads.
1745 if (td_random(td) || !td->o.zone_skip)
1749 * It is time to switch to a new zone if:
1750 * - zone_bytes == zone_size bytes have already been accessed
1751 * - The last position reached the end of the current zone.
1752 * - For reads with td->o.read_beyond_wp == false, the last position
1753 * reached the zone write pointer.
1755 if (td->zone_bytes >= td->o.zone_size ||
1756 f->last_pos[ddir] >= zbd_zone_end(z) ||
1757 (ddir == DDIR_READ &&
1758 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1763 f->file_offset += td->o.zone_size + td->o.zone_skip;
1766 * Wrap from the beginning, if we exceed the file size
1768 if (f->file_offset >= f->real_file_size)
1769 f->file_offset = get_start_offset(td, f);
1771 f->last_pos[ddir] = f->file_offset;
1772 td->io_skip_bytes += td->o.zone_skip;
1777 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1779 * @td: FIO thread data.
1780 * @io_u: FIO I/O unit.
1781 * @ddir: I/O direction before adjustment.
1783 * Return adjusted I/O direction.
1785 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1789 * In case read direction is chosen for the first random I/O, fio with
1790 * zonemode=zbd stops because no data can be read from zoned block
1791 * devices with all empty zones. Overwrite the first I/O direction as
1792 * write to make sure data to read exists.
1794 assert(io_u->file->zbd_info);
1795 if (ddir != DDIR_READ || !td_rw(td))
1798 if (io_u->file->zbd_info->sectors_with_data ||
1799 td->o.read_beyond_wp)
1806 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1807 * @td: FIO thread data.
1808 * @io_u: FIO I/O unit.
1810 * Locking strategy: returns with z->mutex locked if and only if z refers
1811 * to a sequential zone and if io_u_accept is returned. z is the zone that
1812 * corresponds to io_u->offset at the end of this function.
1814 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1816 struct fio_file *f = io_u->file;
1817 struct zoned_block_device_info *zbdi = f->zbd_info;
1818 struct fio_zone_info *zb, *zl, *orig_zb;
1819 uint32_t orig_len = io_u->buflen;
1820 uint64_t min_bs = td->o.min_bs[io_u->ddir];
1826 assert(is_valid_offset(f, io_u->offset));
1827 assert(io_u->buflen);
1829 zb = zbd_offset_to_zone(f, io_u->offset);
1833 /* Accept non-write I/Os for conventional zones. */
1834 if (io_u->ddir != DDIR_WRITE)
1838 * Make sure that writes to conventional zones
1839 * don't cross over to any sequential zones.
1841 if (!(zb + 1)->has_wp ||
1842 io_u->offset + io_u->buflen <= (zb + 1)->start)
1845 if (io_u->offset + min_bs > (zb + 1)->start) {
1847 "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n",
1848 f->file_name, io_u->offset,
1849 min_bs, (zb + 1)->start);
1851 zb->start + (zb + 1)->start - io_u->offset;
1852 new_len = min(io_u->buflen,
1853 (zb + 1)->start - io_u->offset);
1855 new_len = (zb + 1)->start - io_u->offset;
1858 io_u->buflen = new_len / min_bs * min_bs;
1864 * Accept the I/O offset for reads if reading beyond the write pointer
1867 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1868 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1871 zbd_check_swd(td, f);
1873 zone_lock(td, f, zb);
1875 switch (io_u->ddir) {
1877 if (td->runstate == TD_VERIFYING && td_write(td)) {
1878 zb = zbd_replay_write_order(td, io_u, zb);
1883 * Check that there is enough written data in the zone to do an
1884 * I/O of at least min_bs B. If there isn't, find a new zone for
1887 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1888 zb->wp - zb->start : 0;
1889 if (range < min_bs ||
1890 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1892 zl = zbd_get_zone(f, f->max_zone);
1893 zb = zbd_find_zone(td, io_u, min_bs, zb, zl);
1896 "%s: zbd_find_zone(%lld, %llu) failed\n",
1897 f->file_name, io_u->offset,
1902 * zbd_find_zone() returned a zone with a range of at
1905 range = zb->wp - zb->start;
1906 assert(range >= min_bs);
1909 io_u->offset = zb->start;
1913 * Make sure the I/O is within the zone valid data range while
1914 * maximizing the I/O size and preserving randomness.
1916 if (range <= io_u->buflen)
1917 io_u->offset = zb->start;
1918 else if (td_random(td))
1919 io_u->offset = zb->start +
1920 ((io_u->offset - orig_zb->start) %
1921 (range - io_u->buflen)) / min_bs * min_bs;
1924 * When zbd_find_zone() returns a conventional zone,
1925 * we can simply accept the new i/o offset here.
1931 * Make sure the I/O does not cross over the zone wp position.
1933 new_len = min((unsigned long long)io_u->buflen,
1934 (unsigned long long)(zb->wp - io_u->offset));
1935 new_len = new_len / min_bs * min_bs;
1936 if (new_len < io_u->buflen) {
1937 io_u->buflen = new_len;
1938 dprint(FD_IO, "Changed length from %u into %llu\n",
1939 orig_len, io_u->buflen);
1942 assert(zb->start <= io_u->offset);
1943 assert(io_u->offset + io_u->buflen <= zb->wp);
1948 if (io_u->buflen > zbdi->zone_size) {
1949 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
1951 "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n",
1952 f->file_name, io_u->buflen, zbdi->zone_size);
1956 if (!zbd_open_zone(td, f, zb)) {
1958 zb = zbd_convert_to_open_zone(td, io_u);
1960 dprint(FD_IO, "%s: can't convert to open zone",
1966 /* Check whether the zone reset threshold has been exceeded */
1967 if (td->o.zrf.u.f) {
1968 if (zbdi->wp_sectors_with_data >= f->io_size * td->o.zrt.u.f &&
1969 zbd_dec_and_reset_write_cnt(td, f))
1973 /* Reset the zone pointer if necessary */
1974 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
1975 assert(td->o.verify == VERIFY_NONE);
1977 * Since previous write requests may have been submitted
1978 * asynchronously and since we will submit the zone
1979 * reset synchronously, wait until previously submitted
1980 * write requests have completed before issuing a
1985 if (zbd_reset_zone(td, f, zb) < 0)
1988 if (zb->capacity < min_bs) {
1989 td_verror(td, EINVAL, "ZCAP is less min_bs");
1990 log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n",
1991 zb->capacity, min_bs);
1996 /* Make writes occur at the write pointer */
1997 assert(!zbd_zone_full(f, zb, min_bs));
1998 io_u->offset = zb->wp;
1999 if (!is_valid_offset(f, io_u->offset)) {
2000 td_verror(td, EINVAL, "invalid WP value");
2001 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
2002 f->file_name, io_u->offset);
2007 * Make sure that the buflen is a multiple of the minimal
2008 * block size. Give up if shrinking would make the request too
2011 new_len = min((unsigned long long)io_u->buflen,
2012 zbd_zone_capacity_end(zb) - io_u->offset);
2013 new_len = new_len / min_bs * min_bs;
2014 if (new_len == io_u->buflen)
2016 if (new_len >= min_bs) {
2017 io_u->buflen = new_len;
2018 dprint(FD_IO, "Changed length from %u into %llu\n",
2019 orig_len, io_u->buflen);
2023 td_verror(td, EIO, "zone remainder too small");
2024 log_err("zone remainder %lld smaller than min block size %"PRIu64"\n",
2025 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
2030 /* Check random trim targets a non-empty zone */
2031 if (!td_random(td) || zb->wp > zb->start)
2034 /* Find out a non-empty zone to trim */
2036 zl = zbd_get_zone(f, f->max_zone);
2037 zb = zbd_find_zone(td, io_u, 1, zb, zl);
2039 io_u->offset = zb->start;
2040 dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n",
2041 f->file_name, io_u->offset);
2050 case DDIR_SYNC_FILE_RANGE:
2061 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
2062 assert(!io_u->zbd_queue_io);
2063 assert(!io_u->zbd_put_io);
2065 io_u->zbd_queue_io = zbd_queue_io;
2066 io_u->zbd_put_io = zbd_put_io;
2069 * Since we return with the zone lock still held,
2070 * add an annotation to let Coverity know that it
2073 /* coverity[missing_unlock] */
2078 if (zb && zb->has_wp)
2084 /* Return a string with ZBD statistics */
2085 char *zbd_write_status(const struct thread_stat *ts)
2089 if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0)
2095 * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim
2097 * @td: FIO thread data.
2098 * @io_u: FIO I/O unit.
2100 * It is assumed that z->mutex is already locked.
2101 * Return io_u_completed when reset zone succeeds. Return 0 when the target zone
2102 * does not have write pointer. On error, return negative errno.
2104 int zbd_do_io_u_trim(const struct thread_data *td, struct io_u *io_u)
2106 struct fio_file *f = io_u->file;
2107 struct fio_zone_info *z;
2110 z = zbd_offset_to_zone(f, io_u->offset);
2114 if (io_u->offset != z->start) {
2115 log_err("Trim offset not at zone start (%lld)\n",
2120 ret = zbd_reset_zone((struct thread_data *)td, f, z);
2124 return io_u_completed;