X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=zbd.c;h=627fb968ecf78d3fa5e48480c95a2199ae5187c2;hp=fb81b532bc37b7f9d3406442d80f24f7d8e056d9;hb=HEAD;hpb=4d37720ae029bf66556e4f847628f8cb9f35ef47 diff --git a/zbd.c b/zbd.c index fb81b532..8a092cbe 100644 --- a/zbd.c +++ b/zbd.c @@ -7,27 +7,41 @@ #include #include #include -#include #include -#include #include #include -#include + +#include "compiler/compiler.h" +#include "os/os.h" #include "file.h" #include "fio.h" #include "lib/pow2.h" #include "log.h" +#include "oslib/asprintf.h" #include "smalloc.h" #include "verify.h" +#include "pshared.h" #include "zbd.h" +static bool is_valid_offset(const struct fio_file *f, uint64_t offset) +{ + return (uint64_t)(offset - f->file_offset) < f->io_size; +} + +static inline unsigned int zbd_zone_idx(const struct fio_file *f, + struct fio_zone_info *zone) +{ + return zone - f->zbd_info->zone_info; +} + /** - * zbd_zone_idx - convert an offset into a zone number + * zbd_offset_to_zone_idx - convert an offset into a zone number * @f: file pointer. * @offset: offset in bytes. If this offset is in the first zone_size bytes * past the disk size then the index of the sentinel is returned. */ -static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset) +static unsigned int zbd_offset_to_zone_idx(const struct fio_file *f, + uint64_t offset) { uint32_t zone_idx; @@ -39,6 +53,37 @@ static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset) return min(zone_idx, f->zbd_info->nr_zones); } +/** + * zbd_zone_end - Return zone end location + * @z: zone info pointer. + */ +static inline uint64_t zbd_zone_end(const struct fio_zone_info *z) +{ + return (z+1)->start; +} + +/** + * zbd_zone_capacity_end - Return zone capacity limit end location + * @z: zone info pointer. + */ +static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z) +{ + return z->start + z->capacity; +} + +/** + * zbd_zone_remainder - Return the number of bytes that are still available for + * writing before the zone gets full + * @z: zone info pointer. + */ +static inline uint64_t zbd_zone_remainder(struct fio_zone_info *z) +{ + if (z->wp >= zbd_zone_capacity_end(z)) + return 0; + + return zbd_zone_capacity_end(z) - z->wp; +} + /** * zbd_zone_full - verify whether a minimum number of bytes remain in a zone * @f: file pointer. @@ -52,255 +97,672 @@ static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z, { assert((required & 511) == 0); - return z->type == BLK_ZONE_TYPE_SEQWRITE_REQ && - z->wp + required > z->start + f->zbd_info->zone_size; + return z->has_wp && required > zbd_zone_remainder(z); } -static bool is_valid_offset(const struct fio_file *f, uint64_t offset) +static void zone_lock(struct thread_data *td, const struct fio_file *f, + struct fio_zone_info *z) { - return (uint64_t)(offset - f->file_offset) < f->io_size; +#ifndef NDEBUG + unsigned int const nz = zbd_zone_idx(f, z); + /* A thread should never lock zones outside its working area. */ + assert(f->min_zone <= nz && nz < f->max_zone); + assert(z->has_wp); +#endif + + /* + * Lock the io_u target zone. The zone will be unlocked if io_u offset + * is changed or when io_u completes and zbd_put_io() executed. + * To avoid multiple jobs doing asynchronous I/Os from deadlocking each + * other waiting for zone locks when building an io_u batch, first + * only trylock the zone. If the zone is already locked by another job, + * process the currently queued I/Os so that I/O progress is made and + * zones unlocked. + */ + if (pthread_mutex_trylock(&z->mutex) != 0) { + if (!td_ioengine_flagged(td, FIO_SYNCIO)) + io_u_quiesce(td); + pthread_mutex_lock(&z->mutex); + } +} + +static inline void zone_unlock(struct fio_zone_info *z) +{ + assert(z->has_wp); + pthread_mutex_unlock(&z->mutex); } -/* Verify whether direct I/O is used for all host-managed zoned drives. */ +static inline struct fio_zone_info *zbd_get_zone(const struct fio_file *f, + unsigned int zone_idx) +{ + return &f->zbd_info->zone_info[zone_idx]; +} + +static inline struct fio_zone_info * +zbd_offset_to_zone(const struct fio_file *f, uint64_t offset) +{ + return zbd_get_zone(f, zbd_offset_to_zone_idx(f, offset)); +} + +static bool accounting_vdb(struct thread_data *td, const struct fio_file *f) +{ + return td->o.zrt.u.f && td_write(td); +} + +/** + * zbd_get_zoned_model - Get a device zoned model + * @td: FIO thread data + * @f: FIO file for which to get model information + */ +static int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f, + enum zbd_zoned_model *model) +{ + int ret; + + if (f->filetype == FIO_TYPE_PIPE) { + log_err("zonemode=zbd does not support pipes\n"); + return -EINVAL; + } + + /* If regular file, always emulate zones inside the file. */ + if (f->filetype == FIO_TYPE_FILE) { + *model = ZBD_NONE; + return 0; + } + + if (td->io_ops && td->io_ops->get_zoned_model) + ret = td->io_ops->get_zoned_model(td, f, model); + else + ret = blkzoned_get_zoned_model(td, f, model); + if (ret < 0) { + td_verror(td, errno, "get zoned model failed"); + log_err("%s: get zoned model failed (%d).\n", + f->file_name, errno); + } + + return ret; +} + +/** + * zbd_report_zones - Get zone information + * @td: FIO thread data. + * @f: FIO file for which to get zone information + * @offset: offset from which to report zones + * @zones: Array of struct zbd_zone + * @nr_zones: Size of @zones array + * + * Get zone information into @zones starting from the zone at offset @offset + * for the device specified by @f. + * + * Returns the number of zones reported upon success and a negative error code + * upon failure. If the zone report is empty, always assume an error (device + * problem) and return -EIO. + */ +static int zbd_report_zones(struct thread_data *td, struct fio_file *f, + uint64_t offset, struct zbd_zone *zones, + unsigned int nr_zones) +{ + int ret; + + if (td->io_ops && td->io_ops->report_zones) + ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones); + else + ret = blkzoned_report_zones(td, f, offset, zones, nr_zones); + if (ret < 0) { + td_verror(td, errno, "report zones failed"); + log_err("%s: report zones from sector %"PRIu64" failed (nr_zones=%d; errno=%d).\n", + f->file_name, offset >> 9, nr_zones, errno); + } else if (ret == 0) { + td_verror(td, errno, "Empty zone report"); + log_err("%s: report zones from sector %"PRIu64" is empty.\n", + f->file_name, offset >> 9); + ret = -EIO; + } + + return ret; +} + +/** + * zbd_reset_wp - reset the write pointer of a range of zones + * @td: FIO thread data. + * @f: FIO file for which to reset zones + * @offset: Starting offset of the first zone to reset + * @length: Length of the range of zones to reset + * + * Reset the write pointer of all zones in the range @offset...@offset+@length. + * Returns 0 upon success and a negative error code upon failure. + */ +static int zbd_reset_wp(struct thread_data *td, struct fio_file *f, + uint64_t offset, uint64_t length) +{ + int ret; + + if (td->io_ops && td->io_ops->reset_wp) + ret = td->io_ops->reset_wp(td, f, offset, length); + else + ret = blkzoned_reset_wp(td, f, offset, length); + if (ret < 0) { + td_verror(td, errno, "resetting wp failed"); + log_err("%s: resetting wp for %"PRIu64" sectors at sector %"PRIu64" failed (%d).\n", + f->file_name, length >> 9, offset >> 9, errno); + } + + return ret; +} + +/** + * __zbd_reset_zone - reset the write pointer of a single zone + * @td: FIO thread data. + * @f: FIO file associated with the disk for which to reset a write pointer. + * @z: Zone to reset. + * + * Returns 0 upon success and a negative error code upon failure. + * + * The caller must hold z->mutex. + */ +static int __zbd_reset_zone(struct thread_data *td, struct fio_file *f, + struct fio_zone_info *z) +{ + uint64_t offset = z->start; + uint64_t length = (z+1)->start - offset; + uint64_t data_in_zone = z->wp - z->start; + int ret = 0; + + if (!data_in_zone) + return 0; + + assert(is_valid_offset(f, offset + length - 1)); + + dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", + f->file_name, zbd_zone_idx(f, z)); + + switch (f->zbd_info->model) { + case ZBD_HOST_AWARE: + case ZBD_HOST_MANAGED: + ret = zbd_reset_wp(td, f, offset, length); + if (ret < 0) + return ret; + break; + default: + break; + } + + if (accounting_vdb(td, f)) { + pthread_mutex_lock(&f->zbd_info->mutex); + f->zbd_info->wp_valid_data_bytes -= data_in_zone; + pthread_mutex_unlock(&f->zbd_info->mutex); + } + + z->wp = z->start; + + td->ts.nr_zone_resets++; + + return ret; +} + +/** + * zbd_write_zone_put - Remove a zone from the write target zones array. + * @td: FIO thread data. + * @f: FIO file that has the write zones array to remove. + * @zone_idx: Index of the zone to remove. + * + * The caller must hold f->zbd_info->mutex. + */ +static void zbd_write_zone_put(struct thread_data *td, const struct fio_file *f, + struct fio_zone_info *z) +{ + uint32_t zi; + + if (!z->write) + return; + + for (zi = 0; zi < f->zbd_info->num_write_zones; zi++) { + if (zbd_get_zone(f, f->zbd_info->write_zones[zi]) == z) + break; + } + if (zi == f->zbd_info->num_write_zones) + return; + + dprint(FD_ZBD, "%s: removing zone %u from write zone array\n", + f->file_name, zbd_zone_idx(f, z)); + + memmove(f->zbd_info->write_zones + zi, + f->zbd_info->write_zones + zi + 1, + (ZBD_MAX_WRITE_ZONES - (zi + 1)) * + sizeof(f->zbd_info->write_zones[0])); + + f->zbd_info->num_write_zones--; + td->num_write_zones--; + z->write = 0; +} + +/** + * zbd_reset_zone - reset the write pointer of a single zone and remove the zone + * from the array of write zones. + * @td: FIO thread data. + * @f: FIO file associated with the disk for which to reset a write pointer. + * @z: Zone to reset. + * + * Returns 0 upon success and a negative error code upon failure. + * + * The caller must hold z->mutex. + */ +static int zbd_reset_zone(struct thread_data *td, struct fio_file *f, + struct fio_zone_info *z) +{ + int ret; + + ret = __zbd_reset_zone(td, f, z); + if (ret) + return ret; + + pthread_mutex_lock(&f->zbd_info->mutex); + zbd_write_zone_put(td, f, z); + pthread_mutex_unlock(&f->zbd_info->mutex); + return 0; +} + +/** + * zbd_finish_zone - finish the specified zone + * @td: FIO thread data. + * @f: FIO file for which to finish a zone + * @z: Zone to finish. + * + * Finish the zone at @offset with open or close status. + */ +static int zbd_finish_zone(struct thread_data *td, struct fio_file *f, + struct fio_zone_info *z) +{ + uint64_t offset = z->start; + uint64_t length = f->zbd_info->zone_size; + int ret = 0; + + switch (f->zbd_info->model) { + case ZBD_HOST_AWARE: + case ZBD_HOST_MANAGED: + if (td->io_ops && td->io_ops->finish_zone) + ret = td->io_ops->finish_zone(td, f, offset, length); + else + ret = blkzoned_finish_zone(td, f, offset, length); + break; + default: + break; + } + + if (ret < 0) { + td_verror(td, errno, "finish zone failed"); + log_err("%s: finish zone at sector %"PRIu64" failed (%d).\n", + f->file_name, offset >> 9, errno); + } else { + z->wp = (z+1)->start; + } + + return ret; +} + +/** + * zbd_reset_zones - Reset a range of zones. + * @td: fio thread data. + * @f: fio file for which to reset zones + * @zb: first zone to reset. + * @ze: first zone not to reset. + * + * Returns 0 upon success and 1 upon failure. + */ +static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, + struct fio_zone_info *const zb, + struct fio_zone_info *const ze) +{ + struct fio_zone_info *z; + const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; + int res = 0; + + if (fio_unlikely(0 == min_bs)) + return 1; + + dprint(FD_ZBD, "%s: examining zones %u .. %u\n", + f->file_name, zbd_zone_idx(f, zb), zbd_zone_idx(f, ze)); + + for (z = zb; z < ze; z++) { + if (!z->has_wp) + continue; + + zone_lock(td, f, z); + + if (z->wp != z->start) { + dprint(FD_ZBD, "%s: resetting zone %u\n", + f->file_name, zbd_zone_idx(f, z)); + if (zbd_reset_zone(td, f, z) < 0) + res = 1; + } + + zone_unlock(z); + } + + return res; +} + +/** + * zbd_get_max_open_zones - Get the maximum number of open zones + * @td: FIO thread data + * @f: FIO file for which to get max open zones + * @max_open_zones: Upon success, result will be stored here. + * + * A @max_open_zones value set to zero means no limit. + * + * Returns 0 upon success and a negative error code upon failure. + */ +static int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f, + unsigned int *max_open_zones) +{ + int ret; + + if (td->io_ops && td->io_ops->get_max_open_zones) + ret = td->io_ops->get_max_open_zones(td, f, max_open_zones); + else + ret = blkzoned_get_max_open_zones(td, f, max_open_zones); + if (ret < 0) { + td_verror(td, errno, "get max open zones failed"); + log_err("%s: get max open zones failed (%d).\n", + f->file_name, errno); + } + + return ret; +} + +/** + * zbd_get_max_active_zones - Get the maximum number of active zones + * @td: FIO thread data + * @f: FIO file for which to get max active zones + * + * Returns max_active_zones limit value of the target file if it is available. + * Otherwise return zero, which means no limit. + */ +static unsigned int zbd_get_max_active_zones(struct thread_data *td, + struct fio_file *f) +{ + unsigned int max_active_zones; + int ret; + + if (td->io_ops && td->io_ops->get_max_active_zones) + ret = td->io_ops->get_max_active_zones(td, f, + &max_active_zones); + else + ret = blkzoned_get_max_active_zones(td, f, &max_active_zones); + if (ret < 0) { + dprint(FD_ZBD, "%s: max_active_zones is not available\n", + f->file_name); + return 0; + } + + return max_active_zones; +} + +/** + * __zbd_write_zone_get - Add a zone to the array of write zones. + * @td: fio thread data. + * @f: fio file that has the write zones array to add. + * @zone_idx: Index of the zone to add. + * + * Do same operation as @zbd_write_zone_get, except it adds the zone at + * @zone_idx to write target zones array even when it does not have remainder + * space to write one block. + */ +static bool __zbd_write_zone_get(struct thread_data *td, + const struct fio_file *f, + struct fio_zone_info *z) +{ + struct zoned_block_device_info *zbdi = f->zbd_info; + uint32_t zone_idx = zbd_zone_idx(f, z); + bool res = true; + + if (z->cond == ZBD_ZONE_COND_OFFLINE) + return false; + + /* + * Skip full zones with data verification enabled because resetting a + * zone causes data loss and hence causes verification to fail. + */ + if (td->o.verify != VERIFY_NONE && zbd_zone_remainder(z) == 0) + return false; + + /* + * zbdi->max_write_zones == 0 means that there is no limit on the + * maximum number of write target zones. In this case, do no track write + * target zones in zbdi->write_zones array. + */ + if (!zbdi->max_write_zones) + return true; + + pthread_mutex_lock(&zbdi->mutex); + + if (z->write) { + /* + * If the zone is going to be completely filled by writes + * already in-flight, handle it as a full zone instead of a + * write target zone. + */ + if (!zbd_zone_remainder(z)) + res = false; + goto out; + } + + res = false; + /* Zero means no limit */ + if (td->o.job_max_open_zones > 0 && + td->num_write_zones >= td->o.job_max_open_zones) + goto out; + if (zbdi->num_write_zones >= zbdi->max_write_zones) + goto out; + + dprint(FD_ZBD, "%s: adding zone %u to write zone array\n", + f->file_name, zone_idx); + + zbdi->write_zones[zbdi->num_write_zones++] = zone_idx; + td->num_write_zones++; + z->write = 1; + res = true; + +out: + pthread_mutex_unlock(&zbdi->mutex); + return res; +} + +/** + * zbd_write_zone_get - Add a zone to the array of write zones. + * @td: fio thread data. + * @f: fio file that has the open zones to add. + * @zone_idx: Index of the zone to add. + * + * Add a ZBD zone to write target zones array, if it is not yet added. Returns + * true if either the zone was already added or if the zone was successfully + * added to the array without exceeding the maximum number of write zones. + * Returns false if the zone was not already added and addition of the zone + * would cause the zone limit to be exceeded. + */ +static bool zbd_write_zone_get(struct thread_data *td, const struct fio_file *f, + struct fio_zone_info *z) +{ + const uint64_t min_bs = td->o.min_bs[DDIR_WRITE]; + + /* + * Skip full zones with data verification enabled because resetting a + * zone causes data loss and hence causes verification to fail. + */ + if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) + return false; + + return __zbd_write_zone_get(td, f, z); +} + +/* Verify whether direct I/O is used for all host-managed zoned block drives. */ static bool zbd_using_direct_io(void) { - struct thread_data *td; struct fio_file *f; - int i, j; + int j; - for_each_td(td, i) { + for_each_td(td) { if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE)) continue; for_each_file(td, f, j) { - if (f->zbd_info && - f->zbd_info->model == ZBD_DM_HOST_MANAGED) + if (f->zbd_info && f->filetype == FIO_TYPE_BLOCK && + f->zbd_info->model == ZBD_HOST_MANAGED) return false; } - } + } end_for_each(); return true; } /* Whether or not the I/O range for f includes one or more sequential zones */ -static bool zbd_is_seq_job(struct fio_file *f) +static bool zbd_is_seq_job(const struct fio_file *f) { uint32_t zone_idx, zone_idx_b, zone_idx_e; assert(f->zbd_info); + if (f->io_size == 0) return false; - zone_idx_b = zbd_zone_idx(f, f->file_offset); - zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1); + + zone_idx_b = zbd_offset_to_zone_idx(f, f->file_offset); + zone_idx_e = + zbd_offset_to_zone_idx(f, f->file_offset + f->io_size - 1); for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++) - if (f->zbd_info->zone_info[zone_idx].type == - BLK_ZONE_TYPE_SEQWRITE_REQ) + if (zbd_get_zone(f, zone_idx)->has_wp) return true; return false; } /* - * Verify whether offset and size parameters are aligned with zone boundaries. + * Verify whether the file offset and size parameters are aligned with zone + * boundaries. If the file offset is not aligned, align it down to the start of + * the zone containing the start offset and align up the file io_size parameter. */ -static bool zbd_verify_sizes(void) +static bool zbd_zone_align_file_sizes(struct thread_data *td, + struct fio_file *f) { const struct fio_zone_info *z; - struct thread_data *td; - struct fio_file *f; uint64_t new_offset, new_end; - uint32_t zone_idx; - int i, j; - - for_each_td(td, i) { - for_each_file(td, f, j) { - if (!f->zbd_info) - continue; - if (f->file_offset >= f->real_file_size) - continue; - if (!zbd_is_seq_job(f)) - continue; - if (td->o.zone_skip && - (td->o.zone_skip < td->o.zone_size || - td->o.zone_skip % td->o.zone_size)) { - log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", - f->file_name, (unsigned long long) td->o.zone_skip, - (unsigned long long) td->o.zone_size); - return false; - } + if (!f->zbd_info) + return true; + if (f->file_offset >= f->real_file_size) + return true; + if (!zbd_is_seq_job(f)) + return true; - zone_idx = zbd_zone_idx(f, f->file_offset); - z = &f->zbd_info->zone_info[zone_idx]; - if (f->file_offset != z->start) { - new_offset = (z+1)->start; - if (new_offset >= f->file_offset + f->io_size) { - log_info("%s: io_size must be at least one zone\n", - f->file_name); - return false; - } - log_info("%s: rounded up offset from %llu to %llu\n", - f->file_name, (unsigned long long) f->file_offset, - (unsigned long long) new_offset); - f->io_size -= (new_offset - f->file_offset); - f->file_offset = new_offset; - } - zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size); - z = &f->zbd_info->zone_info[zone_idx]; - new_end = z->start; - if (f->file_offset + f->io_size != new_end) { - if (new_end <= f->file_offset) { - log_info("%s: io_size must be at least one zone\n", - f->file_name); - return false; - } - log_info("%s: rounded down io_size from %llu to %llu\n", - f->file_name, (unsigned long long) f->io_size, - (unsigned long long) new_end - f->file_offset); - f->io_size = new_end - f->file_offset; - } + if (!td->o.zone_size) { + td->o.zone_size = f->zbd_info->zone_size; + if (!td->o.zone_size) { + log_err("%s: invalid 0 zone size\n", + f->file_name); + return false; } + } else if (td->o.zone_size != f->zbd_info->zone_size) { + log_err("%s: zonesize %llu does not match the device zone size %"PRIu64".\n", + f->file_name, td->o.zone_size, + f->zbd_info->zone_size); + return false; } - return true; -} - -static bool zbd_verify_bs(void) -{ - struct thread_data *td; - struct fio_file *f; - uint32_t zone_size; - int i, j, k; + if (td->o.zone_skip % td->o.zone_size) { + log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n", + f->file_name, td->o.zone_skip, + td->o.zone_size); + return false; + } - for_each_td(td, i) { - for_each_file(td, f, j) { - if (!f->zbd_info) - continue; - zone_size = f->zbd_info->zone_size; - for (k = 0; k < ARRAY_SIZE(td->o.bs); k++) { - if (td->o.verify != VERIFY_NONE && - zone_size % td->o.bs[k] != 0) { - log_info("%s: block size %llu is not a divisor of the zone size %d\n", - f->file_name, td->o.bs[k], - zone_size); - return false; - } - } + if (td->o.td_ddir == TD_DDIR_READ) { + z = zbd_offset_to_zone(f, f->file_offset + f->io_size); + new_end = z->start; + if (f->file_offset + f->io_size > new_end) { + log_info("%s: rounded io_size from %"PRIu64" to %"PRIu64"\n", + f->file_name, f->io_size, + new_end - f->file_offset); + f->io_size = new_end - f->file_offset; } + return true; } - return true; -} - -/* - * Read zone information into @buf starting from sector @start_sector. - * @fd is a file descriptor that refers to a block device and @bufsz is the - * size of @buf. - * - * Returns 0 upon success and a negative error code upon failure. - * If the zone report is empty, always assume an error (device problem) and - * return -EIO. - */ -static int read_zone_info(int fd, uint64_t start_sector, - void *buf, unsigned int bufsz) -{ - struct blk_zone_report *hdr = buf; - int ret; - if (bufsz < sizeof(*hdr)) - return -EINVAL; + z = zbd_offset_to_zone(f, f->file_offset); + if (f->file_offset != z->start) { + new_offset = zbd_zone_end(z); + if (new_offset >= f->file_offset + f->io_size) { + log_info("%s: io_size must be at least one zone\n", + f->file_name); + return false; + } + log_info("%s: rounded up offset from %"PRIu64" to %"PRIu64"\n", + f->file_name, f->file_offset, + new_offset); + f->io_size -= (new_offset - f->file_offset); + f->file_offset = new_offset; + } - memset(hdr, 0, sizeof(*hdr)); + z = zbd_offset_to_zone(f, f->file_offset + f->io_size); + new_end = z->start; + if (f->file_offset + f->io_size != new_end) { + if (new_end <= f->file_offset) { + log_info("%s: io_size must be at least one zone\n", + f->file_name); + return false; + } + log_info("%s: rounded down io_size from %"PRIu64" to %"PRIu64"\n", + f->file_name, f->io_size, + new_end - f->file_offset); + f->io_size = new_end - f->file_offset; + } - hdr->nr_zones = (bufsz - sizeof(*hdr)) / sizeof(struct blk_zone); - hdr->sector = start_sector; - ret = ioctl(fd, BLKREPORTZONE, hdr); - if (ret) - return -errno; - if (!hdr->nr_zones) - return -EIO; - return 0; + return true; } /* - * Read up to 255 characters from the first line of a file. Strip the trailing - * newline. + * Verify whether offset and size parameters are aligned with zone boundaries. */ -static char *read_file(const char *path) +static bool zbd_verify_sizes(void) { - char line[256], *p = line; - FILE *f; + struct fio_file *f; + int j; - f = fopen(path, "rb"); - if (!f) - return NULL; - if (!fgets(line, sizeof(line), f)) - line[0] = '\0'; - strsep(&p, "\n"); - fclose(f); + for_each_td(td) { + for_each_file(td, f, j) { + if (!zbd_zone_align_file_sizes(td, f)) + return false; + } + } end_for_each(); - return strdup(line); + return true; } -static enum blk_zoned_model get_zbd_model(const char *file_name) +static bool zbd_verify_bs(void) { - enum blk_zoned_model model = ZBD_DM_NONE; - char *zoned_attr_path = NULL; - char *model_str = NULL; - struct stat statbuf; - char *sys_devno_path = NULL; - char *part_attr_path = NULL; - char *part_str = NULL; - char sys_path[PATH_MAX]; - ssize_t sz; - char *delim = NULL; - - if (stat(file_name, &statbuf) < 0) - goto out; - - if (asprintf(&sys_devno_path, "/sys/dev/block/%d:%d", - major(statbuf.st_rdev), minor(statbuf.st_rdev)) < 0) - goto out; - - sz = readlink(sys_devno_path, sys_path, sizeof(sys_path) - 1); - if (sz < 0) - goto out; - sys_path[sz] = '\0'; - - /* - * If the device is a partition device, cut the device name in the - * canonical sysfs path to obtain the sysfs path of the holder device. - * e.g.: /sys/devices/.../sda/sda1 -> /sys/devices/.../sda - */ - if (asprintf(&part_attr_path, "/sys/dev/block/%s/partition", - sys_path) < 0) - goto out; - part_str = read_file(part_attr_path); - if (part_str && *part_str == '1') { - delim = strrchr(sys_path, '/'); - if (!delim) - goto out; - *delim = '\0'; - } - - if (asprintf(&zoned_attr_path, - "/sys/dev/block/%s/queue/zoned", sys_path) < 0) - goto out; + struct fio_file *f; + int j; + + for_each_td(td) { + if (td_trim(td) && + (td->o.min_bs[DDIR_TRIM] != td->o.max_bs[DDIR_TRIM] || + td->o.bssplit_nr[DDIR_TRIM])) { + log_info("bsrange and bssplit are not allowed for trim with zonemode=zbd\n"); + return false; + } + for_each_file(td, f, j) { + uint64_t zone_size; - model_str = read_file(zoned_attr_path); - if (!model_str) - goto out; - dprint(FD_ZBD, "%s: zbd model string: %s\n", file_name, model_str); - if (strcmp(model_str, "host-aware") == 0) - model = ZBD_DM_HOST_AWARE; - else if (strcmp(model_str, "host-managed") == 0) - model = ZBD_DM_HOST_MANAGED; + if (!f->zbd_info) + continue; -out: - free(model_str); - free(zoned_attr_path); - free(part_str); - free(part_attr_path); - free(sys_devno_path); - return model; + zone_size = f->zbd_info->zone_size; + if (td_trim(td) && td->o.bs[DDIR_TRIM] != zone_size) { + log_info("%s: trim block size %llu is not the zone size %"PRIu64"\n", + f->file_name, td->o.bs[DDIR_TRIM], + zone_size); + return false; + } + } + } end_for_each(); + return true; } static int ilog2(uint64_t i) @@ -323,12 +785,12 @@ static int init_zone_info(struct thread_data *td, struct fio_file *f) uint32_t nr_zones; struct fio_zone_info *p; uint64_t zone_size = td->o.zone_size; + uint64_t zone_capacity = td->o.zone_capacity; struct zoned_block_device_info *zbd_info = NULL; - pthread_mutexattr_t attr; int i; if (zone_size == 0) { - log_err("%s: Specifying the zone size is mandatory for regular block devices with --zonemode=zbd\n\n", + log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n", f->file_name); return 1; } @@ -339,24 +801,39 @@ static int init_zone_info(struct thread_data *td, struct fio_file *f) return 1; } + if (zone_capacity == 0) + zone_capacity = zone_size; + + if (zone_capacity > zone_size) { + log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n", + f->file_name, td->o.zone_capacity, td->o.zone_size); + return 1; + } + + if (f->real_file_size < zone_size) { + log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n", + f->file_name, f->real_file_size, zone_size); + return -EINVAL; + } + nr_zones = (f->real_file_size + zone_size - 1) / zone_size; zbd_info = scalloc(1, sizeof(*zbd_info) + (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); if (!zbd_info) return -ENOMEM; - pthread_mutexattr_init(&attr); - pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE); - pthread_mutexattr_setpshared(&attr, true); - pthread_mutex_init(&zbd_info->mutex, &attr); + mutex_init_pshared(&zbd_info->mutex); zbd_info->refcount = 1; p = &zbd_info->zone_info[0]; for (i = 0; i < nr_zones; i++, p++) { - pthread_mutex_init(&p->mutex, &attr); + mutex_init_pshared_with_type(&p->mutex, + PTHREAD_MUTEX_RECURSIVE); p->start = i * zone_size; - p->wp = p->start + zone_size; - p->type = BLK_ZONE_TYPE_SEQWRITE_REQ; - p->cond = BLK_ZONE_COND_EMPTY; + p->wp = p->start; + p->type = ZBD_ZONE_TYPE_SWR; + p->cond = ZBD_ZONE_COND_EMPTY; + p->capacity = zone_capacity; + p->has_wp = 1; } /* a sentinel */ p->start = nr_zones * zone_size; @@ -364,138 +841,194 @@ static int init_zone_info(struct thread_data *td, struct fio_file *f) f->zbd_info = zbd_info; f->zbd_info->zone_size = zone_size; f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? - ilog2(zone_size) : -1; + ilog2(zone_size) : 0; f->zbd_info->nr_zones = nr_zones; - pthread_mutexattr_destroy(&attr); return 0; } /* - * Parse the BLKREPORTZONE output and store it in f->zbd_info. Must be called - * only for devices that support this ioctl, namely zoned block devices. + * Maximum number of zones to report in one operation. + */ +#define ZBD_REPORT_MAX_ZONES 8192U + +/* + * Parse the device zone report and store it in f->zbd_info. Must be called + * only for devices that are zoned, namely those with a model != ZBD_NONE. */ static int parse_zone_info(struct thread_data *td, struct fio_file *f) { - const unsigned int bufsz = sizeof(struct blk_zone_report) + - 4096 * sizeof(struct blk_zone); - uint32_t nr_zones; - struct blk_zone_report *hdr; - const struct blk_zone *z; + int nr_zones, nrz; + struct zbd_zone *zones, *z; struct fio_zone_info *p; - uint64_t zone_size, start_sector; + uint64_t zone_size, offset, capacity; + bool same_zone_cap = true; struct zoned_block_device_info *zbd_info = NULL; - pthread_mutexattr_t attr; - void *buf; - int fd, i, j, ret = 0; + int i, j, ret = -ENOMEM; - pthread_mutexattr_init(&attr); - pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE); - pthread_mutexattr_setpshared(&attr, true); - - buf = malloc(bufsz); - if (!buf) + zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone)); + if (!zones) goto out; - fd = open(f->file_name, O_RDONLY | O_LARGEFILE); - if (fd < 0) { - ret = -errno; - goto free; + nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES); + if (nrz < 0) { + ret = nrz; + log_info("fio: report zones (offset 0) failed for %s (%d).\n", + f->file_name, -ret); + goto out; } - ret = read_zone_info(fd, 0, buf, bufsz); - if (ret < 0) { - log_info("fio: BLKREPORTZONE(%lu) failed for %s (%d).\n", - 0UL, f->file_name, -ret); - goto close; - } - hdr = buf; - if (hdr->nr_zones < 1) { - log_info("fio: %s has invalid zone information.\n", - f->file_name); - goto close; - } - z = (void *)(hdr + 1); - zone_size = z->len << 9; + zone_size = zones[0].len; + capacity = zones[0].capacity; nr_zones = (f->real_file_size + zone_size - 1) / zone_size; if (td->o.zone_size == 0) { td->o.zone_size = zone_size; } else if (td->o.zone_size != zone_size) { - log_err("fio: %s job parameter zonesize %llu does not match disk zone size %llu.\n", - f->file_name, (unsigned long long) td->o.zone_size, - (unsigned long long) zone_size); + log_err("fio: %s job parameter zonesize %llu does not match disk zone size %"PRIu64".\n", + f->file_name, td->o.zone_size, zone_size); ret = -EINVAL; - goto close; + goto out; } - dprint(FD_ZBD, "Device %s has %d zones of size %llu KB\n", f->file_name, - nr_zones, (unsigned long long) zone_size / 1024); + dprint(FD_ZBD, "Device %s has %d zones of size %"PRIu64" KB\n", + f->file_name, nr_zones, zone_size / 1024); zbd_info = scalloc(1, sizeof(*zbd_info) + (nr_zones + 1) * sizeof(zbd_info->zone_info[0])); - ret = -ENOMEM; if (!zbd_info) - goto close; - pthread_mutex_init(&zbd_info->mutex, &attr); + goto out; + mutex_init_pshared(&zbd_info->mutex); zbd_info->refcount = 1; p = &zbd_info->zone_info[0]; - for (start_sector = 0, j = 0; j < nr_zones;) { - z = (void *)(hdr + 1); - for (i = 0; i < hdr->nr_zones; i++, j++, z++, p++) { - pthread_mutex_init(&p->mutex, &attr); - p->start = z->start << 9; + for (offset = 0, j = 0; j < nr_zones;) { + z = &zones[0]; + for (i = 0; i < nrz; i++, j++, z++, p++) { + mutex_init_pshared_with_type(&p->mutex, + PTHREAD_MUTEX_RECURSIVE); + p->start = z->start; + p->capacity = z->capacity; + if (capacity != z->capacity) + same_zone_cap = false; + switch (z->cond) { - case BLK_ZONE_COND_NOT_WP: - case BLK_ZONE_COND_FULL: - p->wp = p->start + zone_size; + case ZBD_ZONE_COND_NOT_WP: + case ZBD_ZONE_COND_FULL: + p->wp = p->start + p->capacity; break; default: assert(z->start <= z->wp); - assert(z->wp <= z->start + (zone_size >> 9)); - p->wp = z->wp << 9; + assert(z->wp <= z->start + zone_size); + p->wp = z->wp; break; } + + switch (z->type) { + case ZBD_ZONE_TYPE_SWR: + p->has_wp = 1; + break; + default: + p->has_wp = 0; + } p->type = z->type; p->cond = z->cond; + if (j > 0 && p->start != p[-1].start + zone_size) { - log_info("%s: invalid zone data\n", - f->file_name); + log_info("%s: invalid zone data [%d:%d]: %"PRIu64" + %"PRIu64" != %"PRIu64"\n", + f->file_name, j, i, + p[-1].start, zone_size, p->start); ret = -EINVAL; - goto close; + goto out; } } z--; - start_sector = z->start + z->len; + offset = z->start + z->len; if (j >= nr_zones) break; - ret = read_zone_info(fd, start_sector, buf, bufsz); - if (ret < 0) { - log_info("fio: BLKREPORTZONE(%llu) failed for %s (%d).\n", - (unsigned long long) start_sector, f->file_name, -ret); - goto close; + + nrz = zbd_report_zones(td, f, offset, zones, + min((uint32_t)(nr_zones - j), + ZBD_REPORT_MAX_ZONES)); + if (nrz < 0) { + ret = nrz; + log_info("fio: report zones (offset %"PRIu64") failed for %s (%d).\n", + offset, f->file_name, -ret); + goto out; } } + /* a sentinel */ - zbd_info->zone_info[nr_zones].start = start_sector << 9; + zbd_info->zone_info[nr_zones].start = offset; f->zbd_info = zbd_info; f->zbd_info->zone_size = zone_size; f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ? - ilog2(zone_size) : -1; + ilog2(zone_size) : 0; f->zbd_info->nr_zones = nr_zones; + f->zbd_info->max_active_zones = zbd_get_max_active_zones(td, f); + + if (same_zone_cap) + dprint(FD_ZBD, "Zone capacity = %"PRIu64" KB\n", + capacity / 1024); + zbd_info = NULL; ret = 0; -close: - sfree(zbd_info); - close(fd); -free: - free(buf); out: - pthread_mutexattr_destroy(&attr); + sfree(zbd_info); + free(zones); return ret; } +static int zbd_set_max_write_zones(struct thread_data *td, struct fio_file *f) +{ + struct zoned_block_device_info *zbd = f->zbd_info; + unsigned int max_open_zones; + int ret; + + if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) { + /* Only host-managed devices have a max open limit */ + zbd->max_write_zones = td->o.max_open_zones; + goto out; + } + + /* If host-managed, get the max open limit */ + ret = zbd_get_max_open_zones(td, f, &max_open_zones); + if (ret) + return ret; + + if (!max_open_zones) { + /* No device limit */ + zbd->max_write_zones = td->o.max_open_zones; + } else if (!td->o.max_open_zones) { + /* No user limit. Set limit to device limit */ + zbd->max_write_zones = max_open_zones; + } else if (td->o.max_open_zones <= max_open_zones) { + /* Both user limit and dev limit. User limit not too large */ + zbd->max_write_zones = td->o.max_open_zones; + } else { + /* Both user limit and dev limit. User limit too large */ + td_verror(td, EINVAL, + "Specified --max_open_zones is too large"); + log_err("Specified --max_open_zones (%d) is larger than max (%u)\n", + td->o.max_open_zones, max_open_zones); + return -EINVAL; + } + +out: + /* Ensure that the limit is not larger than FIO's internal limit */ + if (zbd->max_write_zones > ZBD_MAX_WRITE_ZONES) { + td_verror(td, EINVAL, "'max_open_zones' value is too large"); + log_err("'max_open_zones' value is larger than %u\n", + ZBD_MAX_WRITE_ZONES); + return -EINVAL; + } + + dprint(FD_ZBD, "%s: using max write zones limit: %"PRIu32"\n", + f->file_name, zbd->max_write_zones); + + return 0; +} + /* * Allocate zone information and store it into f->zbd_info if zonemode=zbd. * @@ -503,32 +1036,50 @@ out: */ static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f) { - enum blk_zoned_model zbd_model; - int ret = 0; + enum zbd_zoned_model zbd_model; + int ret; assert(td->o.zone_mode == ZONE_MODE_ZBD); - zbd_model = get_zbd_model(f->file_name); + ret = zbd_get_zoned_model(td, f, &zbd_model); + if (ret) + return ret; + switch (zbd_model) { - case ZBD_DM_HOST_AWARE: - case ZBD_DM_HOST_MANAGED: + case ZBD_HOST_AWARE: + case ZBD_HOST_MANAGED: ret = parse_zone_info(td, f); + if (ret) + return ret; break; - case ZBD_DM_NONE: + case ZBD_NONE: ret = init_zone_info(td, f); + if (ret) + return ret; break; + default: + td_verror(td, EINVAL, "Unsupported zoned model"); + log_err("Unsupported zoned model\n"); + return -EINVAL; } - if (ret == 0) - f->zbd_info->model = zbd_model; - return ret; + + assert(f->zbd_info); + f->zbd_info->model = zbd_model; + + ret = zbd_set_max_write_zones(td, f); + if (ret) { + zbd_free_zone_info(f); + return ret; + } + + return 0; } void zbd_free_zone_info(struct fio_file *f) { uint32_t refcount; - if (!f->zbd_info) - return; + assert(f->zbd_info); pthread_mutex_lock(&f->zbd_info->mutex); refcount = --f->zbd_info->refcount; @@ -550,11 +1101,10 @@ void zbd_free_zone_info(struct fio_file *f) */ static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) { - struct thread_data *td2; struct fio_file *f2; - int i, j, ret; + int j, ret; - for_each_td(td2, i) { + for_each_td(td2) { for_each_file(td2, f2, j) { if (td2 == td && f2 == file) continue; @@ -565,26 +1115,107 @@ static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file) file->zbd_info->refcount++; return 0; } - } + } end_for_each(); ret = zbd_create_zone_info(td, file); if (ret < 0) td_verror(td, -ret, "zbd_create_zone_info() failed"); + return ret; } -int zbd_init(struct thread_data *td) +int zbd_init_files(struct thread_data *td) { struct fio_file *f; int i; for_each_file(td, f, i) { - if (f->filetype != FIO_TYPE_BLOCK) - continue; if (zbd_init_zone_info(td, f)) return 1; } + return 0; +} + +void zbd_recalc_options_with_zone_granularity(struct thread_data *td) +{ + struct fio_file *f; + int i; + + for_each_file(td, f, i) { + struct zoned_block_device_info *zbd = f->zbd_info; + uint64_t zone_size; + + /* zonemode=strided doesn't get per-file zone size. */ + zone_size = zbd ? zbd->zone_size : td->o.zone_size; + if (zone_size == 0) + continue; + + if (td->o.size_nz > 0) + td->o.size = td->o.size_nz * zone_size; + if (td->o.io_size_nz > 0) + td->o.io_size = td->o.io_size_nz * zone_size; + if (td->o.start_offset_nz > 0) + td->o.start_offset = td->o.start_offset_nz * zone_size; + if (td->o.offset_increment_nz > 0) + td->o.offset_increment = + td->o.offset_increment_nz * zone_size; + if (td->o.zone_skip_nz > 0) + td->o.zone_skip = td->o.zone_skip_nz * zone_size; + } +} + +static uint64_t zbd_verify_and_set_vdb(struct thread_data *td, + const struct fio_file *f) +{ + struct fio_zone_info *zb, *ze, *z; + uint64_t wp_vdb = 0; + struct zoned_block_device_info *zbdi = f->zbd_info; + + assert(td->runstate < TD_RUNNING); + assert(zbdi); + + if (!accounting_vdb(td, f)) + return 0; + + /* + * Ensure that the I/O range includes one or more sequential zones so + * that f->min_zone and f->max_zone have different values. + */ + if (!zbd_is_seq_job(f)) + return 0; + + if (zbdi->write_min_zone != zbdi->write_max_zone) { + if (zbdi->write_min_zone != f->min_zone || + zbdi->write_max_zone != f->max_zone) { + td_verror(td, EINVAL, + "multi-jobs with different write ranges are " + "not supported with zone_reset_threshold"); + log_err("multi-jobs with different write ranges are " + "not supported with zone_reset_threshold\n"); + } + return 0; + } + + zbdi->write_min_zone = f->min_zone; + zbdi->write_max_zone = f->max_zone; + + zb = zbd_get_zone(f, f->min_zone); + ze = zbd_get_zone(f, f->max_zone); + for (z = zb; z < ze; z++) + if (z->has_wp) + wp_vdb += z->wp - z->start; + + zbdi->wp_valid_data_bytes = wp_vdb; + + return wp_vdb; +} + +int zbd_setup_files(struct thread_data *td) +{ + struct fio_file *f; + int i; + if (!zbd_using_direct_io()) { log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n"); return 1; @@ -596,169 +1227,102 @@ int zbd_init(struct thread_data *td) if (!zbd_verify_bs()) return 1; - return 0; -} + if (td->o.experimental_verify) { + log_err("zonemode=zbd does not support experimental verify\n"); + return 1; + } -/** - * zbd_reset_range - reset zones for a range of sectors - * @td: FIO thread data. - * @f: Fio file for which to reset zones - * @sector: Starting sector in units of 512 bytes - * @nr_sectors: Number of sectors in units of 512 bytes - * - * Returns 0 upon success and a negative error code upon failure. - */ -static int zbd_reset_range(struct thread_data *td, const struct fio_file *f, - uint64_t offset, uint64_t length) -{ - struct blk_zone_range zr = { - .sector = offset >> 9, - .nr_sectors = length >> 9, - }; - uint32_t zone_idx_b, zone_idx_e; - struct fio_zone_info *zb, *ze, *z; - int ret = 0; + for_each_file(td, f, i) { + struct zoned_block_device_info *zbd = f->zbd_info; + struct fio_zone_info *z; + int zi; + uint64_t vdb; - assert(f->fd != -1); - assert(is_valid_offset(f, offset + length - 1)); - switch (f->zbd_info->model) { - case ZBD_DM_HOST_AWARE: - case ZBD_DM_HOST_MANAGED: - ret = ioctl(f->fd, BLKRESETZONE, &zr); - if (ret < 0) { - td_verror(td, errno, "resetting wp failed"); - log_err("%s: resetting wp for %llu sectors at sector %llu failed (%d).\n", - f->file_name, zr.nr_sectors, zr.sector, errno); - return ret; - } - break; - case ZBD_DM_NONE: - break; - } + assert(zbd); - zone_idx_b = zbd_zone_idx(f, offset); - zb = &f->zbd_info->zone_info[zone_idx_b]; - zone_idx_e = zbd_zone_idx(f, offset + length); - ze = &f->zbd_info->zone_info[zone_idx_e]; - for (z = zb; z < ze; z++) { - pthread_mutex_lock(&z->mutex); - pthread_mutex_lock(&f->zbd_info->mutex); - f->zbd_info->sectors_with_data -= z->wp - z->start; - pthread_mutex_unlock(&f->zbd_info->mutex); - z->wp = z->start; - z->verify_block = 0; - pthread_mutex_unlock(&z->mutex); - } + f->min_zone = zbd_offset_to_zone_idx(f, f->file_offset); + f->max_zone = + zbd_offset_to_zone_idx(f, f->file_offset + f->io_size); - td->ts.nr_zone_resets += ze - zb; + vdb = zbd_verify_and_set_vdb(td, f); - return ret; -} + dprint(FD_ZBD, "%s(%s): valid data bytes = %" PRIu64 "\n", + __func__, f->file_name, vdb); -static unsigned int zbd_zone_nr(struct zoned_block_device_info *zbd_info, - struct fio_zone_info *zone) -{ - return zone - zbd_info->zone_info; -} + /* + * When all zones in the I/O range are conventional, io_size + * can be smaller than zone size, making min_zone the same + * as max_zone. This is why the assert below needs to be made + * conditional. + */ + if (zbd_is_seq_job(f)) + assert(f->min_zone < f->max_zone); -/** - * zbd_reset_zone - reset the write pointer of a single zone - * @td: FIO thread data. - * @f: FIO file associated with the disk for which to reset a write pointer. - * @z: Zone to reset. - * - * Returns 0 upon success and a negative error code upon failure. - */ -static int zbd_reset_zone(struct thread_data *td, const struct fio_file *f, - struct fio_zone_info *z) -{ - dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", f->file_name, - zbd_zone_nr(f->zbd_info, z)); + if (td->o.max_open_zones > 0 && + zbd->max_write_zones != td->o.max_open_zones) { + log_err("Different 'max_open_zones' values\n"); + return 1; + } - return zbd_reset_range(td, f, z->start, (z+1)->start - z->start); -} + /* + * The per job max open zones limit cannot be used without a + * global max open zones limit. (As the tracking of open zones + * is disabled when there is no global max open zones limit.) + */ + if (td->o.job_max_open_zones && !zbd->max_write_zones) { + log_err("'job_max_open_zones' cannot be used without a global open zones limit\n"); + return 1; + } -/* - * Reset a range of zones. Returns 0 upon success and 1 upon failure. - * @td: fio thread data. - * @f: fio file for which to reset zones - * @zb: first zone to reset. - * @ze: first zone not to reset. - * @all_zones: whether to reset all zones or only those zones for which the - * write pointer is not a multiple of td->o.min_bs[DDIR_WRITE]. - */ -static int zbd_reset_zones(struct thread_data *td, struct fio_file *f, - struct fio_zone_info *const zb, - struct fio_zone_info *const ze, bool all_zones) -{ - struct fio_zone_info *z, *start_z = ze; - const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; - bool reset_wp; - int res = 0; + /* + * zbd->max_write_zones is the global limit shared for all jobs + * that target the same zoned block device. Force sync the per + * thread global limit with the actual global limit. (The real + * per thread/job limit is stored in td->o.job_max_open_zones). + */ + td->o.max_open_zones = zbd->max_write_zones; - dprint(FD_ZBD, "%s: examining zones %u .. %u\n", f->file_name, - zbd_zone_nr(f->zbd_info, zb), zbd_zone_nr(f->zbd_info, ze)); - assert(f->fd != -1); - for (z = zb; z < ze; z++) { - pthread_mutex_lock(&z->mutex); - switch (z->type) { - case BLK_ZONE_TYPE_SEQWRITE_REQ: - reset_wp = all_zones ? z->wp != z->start : - (td->o.td_ddir & TD_DDIR_WRITE) && - z->wp % min_bs != 0; - if (start_z == ze && reset_wp) { - start_z = z; - } else if (start_z < ze && !reset_wp) { - dprint(FD_ZBD, - "%s: resetting zones %u .. %u\n", - f->file_name, - zbd_zone_nr(f->zbd_info, start_z), - zbd_zone_nr(f->zbd_info, z)); - if (zbd_reset_range(td, f, start_z->start, - z->start - start_z->start) < 0) - res = 1; - start_z = ze; - } - break; - default: - if (start_z == ze) - break; - dprint(FD_ZBD, "%s: resetting zones %u .. %u\n", - f->file_name, zbd_zone_nr(f->zbd_info, start_z), - zbd_zone_nr(f->zbd_info, z)); - if (zbd_reset_range(td, f, start_z->start, - z->start - start_z->start) < 0) - res = 1; - start_z = ze; - break; + for (zi = f->min_zone; zi < f->max_zone; zi++) { + z = &zbd->zone_info[zi]; + if (z->cond != ZBD_ZONE_COND_IMP_OPEN && + z->cond != ZBD_ZONE_COND_EXP_OPEN && + z->cond != ZBD_ZONE_COND_CLOSED) + continue; + if (!zbd->max_active_zones && + z->cond == ZBD_ZONE_COND_CLOSED) + continue; + if (__zbd_write_zone_get(td, f, z)) + continue; + /* + * If the number of open zones exceeds specified limits, + * error out. + */ + log_err("Number of open zones exceeds max_open_zones limit\n"); + return 1; } } - if (start_z < ze) { - dprint(FD_ZBD, "%s: resetting zones %u .. %u\n", f->file_name, - zbd_zone_nr(f->zbd_info, start_z), - zbd_zone_nr(f->zbd_info, z)); - if (zbd_reset_range(td, f, start_z->start, - z->start - start_z->start) < 0) - res = 1; - } - for (z = zb; z < ze; z++) - pthread_mutex_unlock(&z->mutex); - return res; + return 0; } /* * Reset zbd_info.write_cnt, the counter that counts down towards the next * zone reset. */ -static void zbd_reset_write_cnt(const struct thread_data *td, - const struct fio_file *f) +static void _zbd_reset_write_cnt(const struct thread_data *td, + const struct fio_file *f) { assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1); - pthread_mutex_lock(&f->zbd_info->mutex); f->zbd_info->write_cnt = td->o.zrf.u.f ? min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX; +} + +static void zbd_reset_write_cnt(const struct thread_data *td, + const struct fio_file *f) +{ + pthread_mutex_lock(&f->zbd_info->mutex); + _zbd_reset_write_cnt(td, f); pthread_mutex_unlock(&f->zbd_info->mutex); } @@ -772,361 +1336,355 @@ static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td, if (f->zbd_info->write_cnt) write_cnt = --f->zbd_info->write_cnt; if (write_cnt == 0) - zbd_reset_write_cnt(td, f); + _zbd_reset_write_cnt(td, f); pthread_mutex_unlock(&f->zbd_info->mutex); return write_cnt == 0; } -enum swd_action { - CHECK_SWD, - SET_SWD, -}; - -/* Calculate the number of sectors with data (swd) and perform action 'a' */ -static uint64_t zbd_process_swd(const struct fio_file *f, enum swd_action a) -{ - struct fio_zone_info *zb, *ze, *z; - uint64_t swd = 0; - - zb = &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset)]; - ze = &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset + - f->io_size)]; - for (z = zb; z < ze; z++) { - pthread_mutex_lock(&z->mutex); - swd += z->wp - z->start; - } - pthread_mutex_lock(&f->zbd_info->mutex); - switch (a) { - case CHECK_SWD: - assert(f->zbd_info->sectors_with_data == swd); - break; - case SET_SWD: - f->zbd_info->sectors_with_data = swd; - break; - } - pthread_mutex_unlock(&f->zbd_info->mutex); - for (z = zb; z < ze; z++) - pthread_mutex_unlock(&z->mutex); - - return swd; -} - -/* - * The swd check is useful for debugging but takes too much time to leave - * it enabled all the time. Hence it is disabled by default. - */ -static const bool enable_check_swd = false; - -/* Check whether the value of zbd_info.sectors_with_data is correct. */ -static void zbd_check_swd(const struct fio_file *f) -{ - if (!enable_check_swd) - return; - - zbd_process_swd(f, CHECK_SWD); -} - -static void zbd_init_swd(struct fio_file *f) -{ - uint64_t swd; - - swd = zbd_process_swd(f, SET_SWD); - dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name, - swd); -} - -void zbd_file_reset(struct thread_data *td, struct fio_file *f) -{ - struct fio_zone_info *zb, *ze; - uint32_t zone_idx_e; - - if (!f->zbd_info) - return; - - zb = &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset)]; - zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size); - ze = &f->zbd_info->zone_info[zone_idx_e]; - zbd_init_swd(f); - /* - * If data verification is enabled reset the affected zones before - * writing any data to avoid that a zone reset has to be issued while - * writing data, which causes data loss. - */ - zbd_reset_zones(td, f, zb, ze, td->o.verify != VERIFY_NONE && - (td->o.td_ddir & TD_DDIR_WRITE) && - td->runstate != TD_VERIFYING); - zbd_reset_write_cnt(td, f); -} - -/* The caller must hold f->zbd_info->mutex. */ -static bool is_zone_open(const struct thread_data *td, const struct fio_file *f, - unsigned int zone_idx) -{ - struct zoned_block_device_info *zbdi = f->zbd_info; - int i; - - assert(td->o.max_open_zones <= ARRAY_SIZE(zbdi->open_zones)); - assert(zbdi->num_open_zones <= td->o.max_open_zones); - - for (i = 0; i < zbdi->num_open_zones; i++) - if (zbdi->open_zones[i] == zone_idx) - return true; - - return false; -} - -/* - * Open a ZBD zone if it was not yet open. Returns true if either the zone was - * already open or if opening a new zone is allowed. Returns false if the zone - * was not yet open and opening a new zone would cause the zone limit to be - * exceeded. - */ -static bool zbd_open_zone(struct thread_data *td, const struct io_u *io_u, - uint32_t zone_idx) +void zbd_file_reset(struct thread_data *td, struct fio_file *f) { - const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; - const struct fio_file *f = io_u->file; - struct fio_zone_info *z = &f->zbd_info->zone_info[zone_idx]; - bool res = true; + struct fio_zone_info *zb, *ze; + bool verify_data_left = false; - if (z->cond == BLK_ZONE_COND_OFFLINE) - return false; + if (!f->zbd_info || !td_write(td)) + return; + + zb = zbd_get_zone(f, f->min_zone); + ze = zbd_get_zone(f, f->max_zone); /* - * Skip full zones with data verification enabled because resetting a - * zone causes data loss and hence causes verification to fail. + * If data verification is enabled reset the affected zones before + * writing any data to avoid that a zone reset has to be issued while + * writing data, which causes data loss. */ - if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs)) - return false; - - /* Zero means no limit */ - if (!td->o.max_open_zones) - return true; + if (td->o.verify != VERIFY_NONE) { + verify_data_left = td->runstate == TD_VERIFYING || + td->io_hist_len || td->verify_batch; + if (!verify_data_left) + zbd_reset_zones(td, f, zb, ze); + } - pthread_mutex_lock(&f->zbd_info->mutex); - if (is_zone_open(td, f, zone_idx)) - goto out; - res = false; - if (f->zbd_info->num_open_zones >= td->o.max_open_zones) - goto out; - dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx); - f->zbd_info->open_zones[f->zbd_info->num_open_zones++] = zone_idx; - z->open = 1; - res = true; + zbd_reset_write_cnt(td, f); +} -out: - pthread_mutex_unlock(&f->zbd_info->mutex); - return res; +/* Return random zone index for one of the write target zones. */ +static uint32_t pick_random_zone_idx(const struct fio_file *f, + const struct io_u *io_u) +{ + return (io_u->offset - f->file_offset) * + f->zbd_info->num_write_zones / f->io_size; } -/* The caller must hold f->zbd_info->mutex */ -static void zbd_close_zone(struct thread_data *td, const struct fio_file *f, - unsigned int open_zone_idx) +static bool any_io_in_flight(void) { - uint32_t zone_idx; + for_each_td(td) { + if (td->io_u_in_flight) + return true; + } end_for_each(); - assert(open_zone_idx < f->zbd_info->num_open_zones); - zone_idx = f->zbd_info->open_zones[open_zone_idx]; - memmove(f->zbd_info->open_zones + open_zone_idx, - f->zbd_info->open_zones + open_zone_idx + 1, - (FIO_MAX_OPEN_ZBD_ZONES - (open_zone_idx + 1)) * - sizeof(f->zbd_info->open_zones[0])); - f->zbd_info->num_open_zones--; - f->zbd_info->zone_info[zone_idx].open = 0; + return false; } /* - * Modify the offset of an I/O unit that does not refer to an open zone such - * that it refers to an open zone. Close an open zone and open a new zone if - * necessary. This algorithm can only work correctly if all write pointers are + * Modify the offset of an I/O unit that does not refer to a zone such that + * in write target zones array. Add a zone to or remove a zone from the lsit if + * necessary. The write target zone is searched across sequential zones. + * This algorithm can only work correctly if all write pointers are * a multiple of the fio block size. The caller must neither hold z->mutex * nor f->zbd_info->mutex. Returns with z->mutex held upon success. */ -static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td, - struct io_u *io_u) +static struct fio_zone_info *zbd_convert_to_write_zone(struct thread_data *td, + struct io_u *io_u) { - const uint32_t min_bs = td->o.min_bs[io_u->ddir]; - const struct fio_file *f = io_u->file; + const uint64_t min_bs = td->o.min_bs[io_u->ddir]; + struct fio_file *f = io_u->file; + struct zoned_block_device_info *zbdi = f->zbd_info; struct fio_zone_info *z; - unsigned int open_zone_idx = -1; + unsigned int write_zone_idx = -1; uint32_t zone_idx, new_zone_idx; int i; + bool wait_zone_write; + bool in_flight; + bool should_retry = true; assert(is_valid_offset(f, io_u->offset)); - if (td->o.max_open_zones) { + if (zbdi->max_write_zones || td->o.job_max_open_zones) { /* - * This statement accesses f->zbd_info->open_zones[] on purpose + * This statement accesses zbdi->write_zones[] on purpose * without locking. */ - zone_idx = f->zbd_info->open_zones[(io_u->offset - - f->file_offset) * - f->zbd_info->num_open_zones / f->io_size]; + zone_idx = zbdi->write_zones[pick_random_zone_idx(f, io_u)]; } else { - zone_idx = zbd_zone_idx(f, io_u->offset); + zone_idx = zbd_offset_to_zone_idx(f, io_u->offset); } - dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", + if (zone_idx < f->min_zone) + zone_idx = f->min_zone; + else if (zone_idx >= f->max_zone) + zone_idx = f->max_zone - 1; + + dprint(FD_ZBD, + "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n", __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen); /* - * Since z->mutex is the outer lock and f->zbd_info->mutex the inner + * Since z->mutex is the outer lock and zbdi->mutex the inner * lock it can happen that the state of the zone with index zone_idx - * has changed after 'z' has been assigned and before f->zbd_info->mutex + * has changed after 'z' has been assigned and before zbdi->mutex * has been obtained. Hence the loop. */ for (;;) { - z = &f->zbd_info->zone_info[zone_idx]; + uint32_t tmp_idx; + + z = zbd_get_zone(f, zone_idx); + if (z->has_wp) + zone_lock(td, f, z); + + pthread_mutex_lock(&zbdi->mutex); + + if (z->has_wp) { + if (z->cond != ZBD_ZONE_COND_OFFLINE && + zbdi->max_write_zones == 0 && + td->o.job_max_open_zones == 0) + goto examine_zone; + if (zbdi->num_write_zones == 0) { + dprint(FD_ZBD, "%s(%s): no zone is write target\n", + __func__, f->file_name); + goto choose_other_zone; + } + } - pthread_mutex_lock(&z->mutex); - pthread_mutex_lock(&f->zbd_info->mutex); - if (td->o.max_open_zones == 0) - goto examine_zone; - if (f->zbd_info->num_open_zones == 0) { - pthread_mutex_unlock(&f->zbd_info->mutex); - pthread_mutex_unlock(&z->mutex); - dprint(FD_ZBD, "%s(%s): no zones are open\n", - __func__, f->file_name); - return NULL; + /* + * Array of write target zones is per-device, shared across all + * threads. Start with quasi-random candidate zone. Ignore + * zones which don't belong to thread's offset/size area. + */ + write_zone_idx = pick_random_zone_idx(f, io_u); + assert(!write_zone_idx || + write_zone_idx < zbdi->num_write_zones); + tmp_idx = write_zone_idx; + + for (i = 0; i < zbdi->num_write_zones; i++) { + uint32_t tmpz; + + if (tmp_idx >= zbdi->num_write_zones) + tmp_idx = 0; + tmpz = zbdi->write_zones[tmp_idx]; + if (f->min_zone <= tmpz && tmpz < f->max_zone) { + write_zone_idx = tmp_idx; + goto found_candidate_zone; + } + + tmp_idx++; } - open_zone_idx = (io_u->offset - f->file_offset) * - f->zbd_info->num_open_zones / f->io_size; - assert(open_zone_idx < f->zbd_info->num_open_zones); - new_zone_idx = f->zbd_info->open_zones[open_zone_idx]; + + dprint(FD_ZBD, "%s(%s): no candidate zone\n", + __func__, f->file_name); + + pthread_mutex_unlock(&zbdi->mutex); + + if (z->has_wp) + zone_unlock(z); + + return NULL; + +found_candidate_zone: + new_zone_idx = zbdi->write_zones[write_zone_idx]; if (new_zone_idx == zone_idx) break; zone_idx = new_zone_idx; - pthread_mutex_unlock(&f->zbd_info->mutex); - pthread_mutex_unlock(&z->mutex); + + pthread_mutex_unlock(&zbdi->mutex); + + if (z->has_wp) + zone_unlock(z); } - /* Both z->mutex and f->zbd_info->mutex are held. */ + /* Both z->mutex and zbdi->mutex are held. */ examine_zone: - if (z->wp + min_bs <= (z+1)->start) { - pthread_mutex_unlock(&f->zbd_info->mutex); + if (zbd_zone_remainder(z) >= min_bs) { + pthread_mutex_unlock(&zbdi->mutex); goto out; } - dprint(FD_ZBD, "%s(%s): closing zone %d\n", __func__, f->file_name, - zone_idx); - if (td->o.max_open_zones) - zbd_close_zone(td, f, open_zone_idx); - pthread_mutex_unlock(&f->zbd_info->mutex); + +choose_other_zone: + /* Check if number of write target zones reaches one of limits. */ + wait_zone_write = + zbdi->num_write_zones == f->max_zone - f->min_zone || + (zbdi->max_write_zones && + zbdi->num_write_zones == zbdi->max_write_zones) || + (td->o.job_max_open_zones && + td->num_write_zones == td->o.job_max_open_zones); + + pthread_mutex_unlock(&zbdi->mutex); /* Only z->mutex is held. */ - /* Zone 'z' is full, so try to open a new zone. */ - for (i = f->io_size / f->zbd_info->zone_size; i > 0; i--) { + /* + * When number of write target zones reaches to one of limits, wait for + * zone write completion to one of them before trying a new zone. + */ + if (wait_zone_write) { + dprint(FD_ZBD, + "%s(%s): quiesce to remove a zone from write target zones array\n", + __func__, f->file_name); + io_u_quiesce(td); + } + +retry: + /* Zone 'z' is full, so try to choose a new zone. */ + for (i = f->io_size / zbdi->zone_size; i > 0; i--) { zone_idx++; - pthread_mutex_unlock(&z->mutex); + if (z->has_wp) + zone_unlock(z); z++; if (!is_valid_offset(f, z->start)) { /* Wrap-around. */ - zone_idx = zbd_zone_idx(f, f->file_offset); - z = &f->zbd_info->zone_info[zone_idx]; + zone_idx = f->min_zone; + z = zbd_get_zone(f, zone_idx); } assert(is_valid_offset(f, z->start)); - pthread_mutex_lock(&z->mutex); - if (z->open) + if (!z->has_wp) continue; - if (zbd_open_zone(td, io_u, zone_idx)) + zone_lock(td, f, z); + if (z->write) + continue; + if (zbd_write_zone_get(td, f, z)) goto out; } /* Only z->mutex is held. */ - /* Check whether the write fits in any of the already opened zones. */ - pthread_mutex_lock(&f->zbd_info->mutex); - for (i = 0; i < f->zbd_info->num_open_zones; i++) { - zone_idx = f->zbd_info->open_zones[i]; - pthread_mutex_unlock(&f->zbd_info->mutex); - pthread_mutex_unlock(&z->mutex); + /* Check whether the write fits in any of the write target zones. */ + pthread_mutex_lock(&zbdi->mutex); + for (i = 0; i < zbdi->num_write_zones; i++) { + zone_idx = zbdi->write_zones[i]; + if (zone_idx < f->min_zone || zone_idx >= f->max_zone) + continue; + pthread_mutex_unlock(&zbdi->mutex); + zone_unlock(z); - z = &f->zbd_info->zone_info[zone_idx]; + z = zbd_get_zone(f, zone_idx); - pthread_mutex_lock(&z->mutex); - if (z->wp + min_bs <= (z+1)->start) + zone_lock(td, f, z); + if (zbd_zone_remainder(z) >= min_bs) goto out; - pthread_mutex_lock(&f->zbd_info->mutex); + pthread_mutex_lock(&zbdi->mutex); } - pthread_mutex_unlock(&f->zbd_info->mutex); - pthread_mutex_unlock(&z->mutex); - dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__, - f->file_name); + + /* + * When any I/O is in-flight or when all I/Os in-flight get completed, + * the I/Os might have removed zones from the write target array then + * retry the steps to choose a zone. Before retry, call io_u_quiesce() + * to complete in-flight writes. + */ + in_flight = any_io_in_flight(); + if (in_flight || should_retry) { + dprint(FD_ZBD, + "%s(%s): wait zone write and retry write target zone selection\n", + __func__, f->file_name); + should_retry = in_flight; + pthread_mutex_unlock(&zbdi->mutex); + zone_unlock(z); + io_u_quiesce(td); + zone_lock(td, f, z); + goto retry; + } + + pthread_mutex_unlock(&zbdi->mutex); + + zone_unlock(z); + + dprint(FD_ZBD, "%s(%s): did not choose another write zone\n", + __func__, f->file_name); + return NULL; out: - dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name, - zone_idx); - io_u->offset = z->start; - return z; -} - -/* The caller must hold z->mutex. */ -static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td, - struct io_u *io_u, - struct fio_zone_info *z) -{ - const struct fio_file *f = io_u->file; - const uint32_t min_bs = td->o.min_bs[DDIR_WRITE]; + dprint(FD_ZBD, "%s(%s): returning zone %d\n", + __func__, f->file_name, zone_idx); - if (!zbd_open_zone(td, io_u, z - f->zbd_info->zone_info)) { - pthread_mutex_unlock(&z->mutex); - z = zbd_convert_to_open_zone(td, io_u); - assert(z); - } + io_u->offset = z->start; + assert(z->has_wp); + assert(z->cond != ZBD_ZONE_COND_OFFLINE); - if (z->verify_block * min_bs >= f->zbd_info->zone_size) - log_err("%s: %d * %d >= %llu\n", f->file_name, z->verify_block, - min_bs, (unsigned long long) f->zbd_info->zone_size); - io_u->offset = z->start + z->verify_block++ * min_bs; return z; } /* - * Find another zone for which @io_u fits below the write pointer. Start - * searching in zones @zb + 1 .. @zl and continue searching in zones - * @zf .. @zb - 1. + * Find another zone which has @min_bytes of readable data. Search in zones + * @zb + 1 .. @zl. For random workload, also search in zones @zb - 1 .. @zf. * - * Either returns NULL or returns a zone pointer and holds the mutex for that - * zone. + * Either returns NULL or returns a zone pointer. When the zone has write + * pointer, hold the mutex for the zone. */ static struct fio_zone_info * -zbd_find_zone(struct thread_data *td, struct io_u *io_u, +zbd_find_zone(struct thread_data *td, struct io_u *io_u, uint64_t min_bytes, struct fio_zone_info *zb, struct fio_zone_info *zl) { - const uint32_t min_bs = td->o.min_bs[io_u->ddir]; - const struct fio_file *f = io_u->file; + struct fio_file *f = io_u->file; struct fio_zone_info *z1, *z2; - const struct fio_zone_info *const zf = - &f->zbd_info->zone_info[zbd_zone_idx(f, f->file_offset)]; + const struct fio_zone_info *const zf = zbd_get_zone(f, f->min_zone); /* * Skip to the next non-empty zone in case of sequential I/O and to * the nearest non-empty zone in case of random I/O. */ for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) { - if (z1 < zl && z1->cond != BLK_ZONE_COND_OFFLINE) { - pthread_mutex_lock(&z1->mutex); - if (z1->start + min_bs <= z1->wp) + if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) { + if (z1->has_wp) + zone_lock(td, f, z1); + if (z1->start + min_bytes <= z1->wp) return z1; - pthread_mutex_unlock(&z1->mutex); + if (z1->has_wp) + zone_unlock(z1); } else if (!td_random(td)) { break; } + if (td_random(td) && z2 >= zf && - z2->cond != BLK_ZONE_COND_OFFLINE) { - pthread_mutex_lock(&z2->mutex); - if (z2->start + min_bs <= z2->wp) + z2->cond != ZBD_ZONE_COND_OFFLINE) { + if (z2->has_wp) + zone_lock(td, f, z2); + if (z2->start + min_bytes <= z2->wp) return z2; - pthread_mutex_unlock(&z2->mutex); + if (z2->has_wp) + zone_unlock(z2); } } - dprint(FD_ZBD, "%s: adjusting random read offset failed\n", - f->file_name); + + dprint(FD_ZBD, + "%s: no zone has %"PRIu64" bytes of readable data\n", + f->file_name, min_bytes); + return NULL; } +/** + * zbd_end_zone_io - update zone status at command completion + * @io_u: I/O unit + * @z: zone info pointer + * + * If the write command made the zone full, remove it from the write target + * zones array. + * + * The caller must hold z->mutex. + */ +static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u, + struct fio_zone_info *z) +{ + const struct fio_file *f = io_u->file; + + if (io_u->ddir == DDIR_WRITE && + io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) { + pthread_mutex_lock(&f->zbd_info->mutex); + zbd_write_zone_put(td, f, z); + pthread_mutex_unlock(&f->zbd_info->mutex); + } +} + /** * zbd_queue_io - update the write pointer of a sequential zone * @io_u: I/O unit @@ -1136,56 +1694,53 @@ zbd_find_zone(struct thread_data *td, struct io_u *io_u, * For write and trim operations, update the write pointer of the I/O unit * target zone. */ -static void zbd_queue_io(struct io_u *io_u, int q, bool success) +static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q, + bool success) { const struct fio_file *f = io_u->file; struct zoned_block_device_info *zbd_info = f->zbd_info; struct fio_zone_info *z; - uint32_t zone_idx; uint64_t zone_end; - if (!zbd_info) - return; - - zone_idx = zbd_zone_idx(f, io_u->offset); - assert(zone_idx < zbd_info->nr_zones); - z = &zbd_info->zone_info[zone_idx]; + assert(zbd_info); - if (z->type != BLK_ZONE_TYPE_SEQWRITE_REQ) - return; + z = zbd_offset_to_zone(f, io_u->offset); + assert(z->has_wp); if (!success) goto unlock; dprint(FD_ZBD, "%s: queued I/O (%lld, %llu) for zone %u\n", - f->file_name, io_u->offset, io_u->buflen, zone_idx); + f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z)); switch (io_u->ddir) { case DDIR_WRITE: zone_end = min((uint64_t)(io_u->offset + io_u->buflen), - (z + 1)->start); - pthread_mutex_lock(&zbd_info->mutex); + zbd_zone_capacity_end(z)); + /* * z->wp > zone_end means that one or more I/O errors * have occurred. */ - if (z->wp <= zone_end) - zbd_info->sectors_with_data += zone_end - z->wp; - pthread_mutex_unlock(&zbd_info->mutex); + if (accounting_vdb(td, f) && z->wp <= zone_end) { + pthread_mutex_lock(&zbd_info->mutex); + zbd_info->wp_valid_data_bytes += zone_end - z->wp; + pthread_mutex_unlock(&zbd_info->mutex); + } z->wp = zone_end; break; - case DDIR_TRIM: - assert(z->wp == z->start); - break; default: break; } + if (q == FIO_Q_COMPLETED && !io_u->error) + zbd_end_zone_io(td, io_u, z); + unlock: if (!success || q != FIO_Q_QUEUED) { /* BUSY or COMPLETED: unlock the zone */ - pthread_mutex_unlock(&z->mutex); + zone_unlock(z); io_u->zbd_put_io = NULL; } } @@ -1194,31 +1749,32 @@ unlock: * zbd_put_io - Unlock an I/O unit target zone lock * @io_u: I/O unit */ -static void zbd_put_io(const struct io_u *io_u) +static void zbd_put_io(struct thread_data *td, const struct io_u *io_u) { const struct fio_file *f = io_u->file; - struct zoned_block_device_info *zbd_info = f->zbd_info; struct fio_zone_info *z; - uint32_t zone_idx; - if (!zbd_info) - return; - - zone_idx = zbd_zone_idx(f, io_u->offset); - assert(zone_idx < zbd_info->nr_zones); - z = &zbd_info->zone_info[zone_idx]; + assert(f->zbd_info); - if (z->type != BLK_ZONE_TYPE_SEQWRITE_REQ) - return; + z = zbd_offset_to_zone(f, io_u->offset); + assert(z->has_wp); dprint(FD_ZBD, "%s: terminate I/O (%lld, %llu) for zone %u\n", - f->file_name, io_u->offset, io_u->buflen, zone_idx); + f->file_name, io_u->offset, io_u->buflen, zbd_zone_idx(f, z)); + + zbd_end_zone_io(td, io_u, z); - assert(pthread_mutex_unlock(&z->mutex) == 0); - zbd_check_swd(f); + zone_unlock(z); } +/* + * Windows and MacOS do not define this. + */ +#ifndef EREMOTEIO +#define EREMOTEIO 121 /* POSIX value */ +#endif + bool zbd_unaligned_write(int error_code) { switch (error_code) { @@ -1246,10 +1802,29 @@ void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) struct fio_file *f = io_u->file; enum fio_ddir ddir = io_u->ddir; struct fio_zone_info *z; - uint32_t zone_idx; assert(td->o.zone_mode == ZONE_MODE_ZBD); assert(td->o.zone_size); + assert(f->zbd_info); + + z = zbd_offset_to_zone(f, f->last_pos[ddir]); + + /* + * When the zone capacity is smaller than the zone size and the I/O is + * sequential write, skip to zone end if the latest position is at the + * zone capacity limit. + */ + if (z->capacity < f->zbd_info->zone_size && + !td_random(td) && ddir == DDIR_WRITE && + f->last_pos[ddir] >= zbd_zone_capacity_end(z)) { + dprint(FD_ZBD, + "%s: Jump from zone capacity limit to zone end:" + " (%"PRIu64" -> %"PRIu64") for zone %u (%"PRIu64")\n", + f->file_name, f->last_pos[ddir], + zbd_zone_end(z), zbd_zone_idx(f, z), z->capacity); + td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir]; + f->last_pos[ddir] = zbd_zone_end(z); + } /* * zone_skip is valid only for sequential workloads. @@ -1264,11 +1839,8 @@ void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) * - For reads with td->o.read_beyond_wp == false, the last position * reached the zone write pointer. */ - zone_idx = zbd_zone_idx(f, f->last_pos[ddir]); - z = &f->zbd_info->zone_info[zone_idx]; - if (td->zone_bytes >= td->o.zone_size || - f->last_pos[ddir] >= (z+1)->start || + f->last_pos[ddir] >= zbd_zone_end(z) || (ddir == DDIR_READ && (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) { /* @@ -1288,6 +1860,35 @@ void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) } } +/** + * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd. + * + * @td: FIO thread data. + * @io_u: FIO I/O unit. + * @ddir: I/O direction before adjustment. + * + * Return adjusted I/O direction. + */ +enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u, + enum fio_ddir ddir) +{ + /* + * In case read direction is chosen for the first random I/O, fio with + * zonemode=zbd stops because no data can be read from zoned block + * devices with all empty zones. Overwrite the first I/O direction as + * write to make sure data to read exists. + */ + assert(io_u->file->zbd_info); + if (ddir != DDIR_READ || !td_rw(td)) + return ddir; + + if (io_u->file->last_start[DDIR_WRITE] != -1ULL || + td->o.read_beyond_wp || td->o.rwmix[DDIR_WRITE] == 0) + return DDIR_READ; + + return DDIR_WRITE; +} + /** * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives * @td: FIO thread data. @@ -1299,71 +1900,80 @@ void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u) */ enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) { - const struct fio_file *f = io_u->file; - uint32_t zone_idx_b; + struct fio_file *f = io_u->file; + struct zoned_block_device_info *zbdi = f->zbd_info; struct fio_zone_info *zb, *zl, *orig_zb; uint32_t orig_len = io_u->buflen; - uint32_t min_bs = td->o.min_bs[io_u->ddir]; + uint64_t min_bs = td->o.min_bs[io_u->ddir]; uint64_t new_len; int64_t range; - if (!f->zbd_info) - return io_u_accept; - + assert(zbdi); + assert(min_bs); assert(is_valid_offset(f, io_u->offset)); assert(io_u->buflen); - zone_idx_b = zbd_zone_idx(f, io_u->offset); - zb = &f->zbd_info->zone_info[zone_idx_b]; + + zb = zbd_offset_to_zone(f, io_u->offset); orig_zb = zb; - /* Accept the I/O offset for conventional zones. */ - if (zb->type == BLK_ZONE_TYPE_CONVENTIONAL) + if (!zb->has_wp) { + /* Accept non-write I/Os for conventional zones. */ + if (io_u->ddir != DDIR_WRITE) + return io_u_accept; + + /* + * Make sure that writes to conventional zones + * don't cross over to any sequential zones. + */ + if (!(zb + 1)->has_wp || + io_u->offset + io_u->buflen <= (zb + 1)->start) + return io_u_accept; + + if (io_u->offset + min_bs > (zb + 1)->start) { + dprint(FD_IO, + "%s: off=%llu + min_bs=%"PRIu64" > next zone %"PRIu64"\n", + f->file_name, io_u->offset, + min_bs, (zb + 1)->start); + io_u->offset = + zb->start + (zb + 1)->start - io_u->offset; + new_len = min(io_u->buflen, + (zb + 1)->start - io_u->offset); + } else { + new_len = (zb + 1)->start - io_u->offset; + } + + io_u->buflen = new_len / min_bs * min_bs; + return io_u_accept; + } /* * Accept the I/O offset for reads if reading beyond the write pointer * is enabled. */ - if (zb->cond != BLK_ZONE_COND_OFFLINE && + if (zb->cond != ZBD_ZONE_COND_OFFLINE && io_u->ddir == DDIR_READ && td->o.read_beyond_wp) return io_u_accept; - zbd_check_swd(f); - - /* - * Lock the io_u target zone. The zone will be unlocked if io_u offset - * is changed or when io_u completes and zbd_put_io() executed. - * To avoid multiple jobs doing asynchronous I/Os from deadlocking each - * other waiting for zone locks when building an io_u batch, first - * only trylock the zone. If the zone is already locked by another job, - * process the currently queued I/Os so that I/O progress is made and - * zones unlocked. - */ - if (pthread_mutex_trylock(&zb->mutex) != 0) { - if (!td_ioengine_flagged(td, FIO_SYNCIO)) - io_u_quiesce(td); - pthread_mutex_lock(&zb->mutex); - } + zone_lock(td, f, zb); switch (io_u->ddir) { case DDIR_READ: - if (td->runstate == TD_VERIFYING) { - zb = zbd_replay_write_order(td, io_u, zb); + if (td->runstate == TD_VERIFYING && td_write(td)) goto accept; - } + /* * Check that there is enough written data in the zone to do an * I/O of at least min_bs B. If there isn't, find a new zone for * the I/O. */ - range = zb->cond != BLK_ZONE_COND_OFFLINE ? + range = zb->cond != ZBD_ZONE_COND_OFFLINE ? zb->wp - zb->start : 0; if (range < min_bs || ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) { - pthread_mutex_unlock(&zb->mutex); - zl = &f->zbd_info->zone_info[zbd_zone_idx(f, - f->file_offset + f->io_size)]; - zb = zbd_find_zone(td, io_u, zb, zl); + zone_unlock(zb); + zl = zbd_get_zone(f, f->max_zone); + zb = zbd_find_zone(td, io_u, min_bs, zb, zl); if (!zb) { dprint(FD_ZBD, "%s: zbd_find_zone(%lld, %llu) failed\n", @@ -1381,6 +1991,7 @@ enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) if (!td_random(td)) io_u->offset = zb->start; } + /* * Make sure the I/O is within the zone valid data range while * maximizing the I/O size and preserving randomness. @@ -1391,6 +2002,14 @@ enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) io_u->offset = zb->start + ((io_u->offset - orig_zb->start) % (range - io_u->buflen)) / min_bs * min_bs; + + /* + * When zbd_find_zone() returns a conventional zone, + * we can simply accept the new i/o offset here. + */ + if (!zb->has_wp) + return io_u_accept; + /* * Make sure the I/O does not cross over the zone wp position. */ @@ -1402,30 +2021,85 @@ enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) dprint(FD_IO, "Changed length from %u into %llu\n", orig_len, io_u->buflen); } + assert(zb->start <= io_u->offset); assert(io_u->offset + io_u->buflen <= zb->wp); + goto accept; + case DDIR_WRITE: - if (io_u->buflen > f->zbd_info->zone_size) + if (io_u->buflen > zbdi->zone_size) { + td_verror(td, EINVAL, "I/O buflen exceeds zone size"); + dprint(FD_IO, + "%s: I/O buflen %llu exceeds zone size %"PRIu64"\n", + f->file_name, io_u->buflen, zbdi->zone_size); goto eof; - if (!zbd_open_zone(td, io_u, zone_idx_b)) { - pthread_mutex_unlock(&zb->mutex); - zb = zbd_convert_to_open_zone(td, io_u); - if (!zb) + } + +retry: + if (zbd_zone_remainder(zb) > 0 && + zbd_zone_remainder(zb) < min_bs) { + pthread_mutex_lock(&f->zbd_info->mutex); + zbd_write_zone_put(td, f, zb); + pthread_mutex_unlock(&f->zbd_info->mutex); + dprint(FD_ZBD, + "%s: finish zone %d\n", + f->file_name, zbd_zone_idx(f, zb)); + io_u_quiesce(td); + zbd_finish_zone(td, f, zb); + if (zbd_zone_idx(f, zb) + 1 >= f->max_zone) { + if (!td_random(td)) + goto eof; + } + zone_unlock(zb); + + /* Find the next write pointer zone */ + do { + zb++; + if (zbd_zone_idx(f, zb) >= f->max_zone) + zb = zbd_get_zone(f, f->min_zone); + } while (!zb->has_wp); + + zone_lock(td, f, zb); + } + + if (!zbd_write_zone_get(td, f, zb)) { + zone_unlock(zb); + zb = zbd_convert_to_write_zone(td, io_u); + if (!zb) { + dprint(FD_IO, "%s: can't convert to write target zone", + f->file_name); goto eof; - zone_idx_b = zb - f->zbd_info->zone_info; + } } + + if (zbd_zone_remainder(zb) > 0 && + zbd_zone_remainder(zb) < min_bs) + goto retry; + /* Check whether the zone reset threshold has been exceeded */ if (td->o.zrf.u.f) { - if (f->zbd_info->sectors_with_data >= + if (zbdi->wp_valid_data_bytes >= f->io_size * td->o.zrt.u.f && - zbd_dec_and_reset_write_cnt(td, f)) { + zbd_dec_and_reset_write_cnt(td, f)) zb->reset_zone = 1; - } } + /* Reset the zone pointer if necessary */ if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) { - assert(td->o.verify == VERIFY_NONE); + if (td->o.verify != VERIFY_NONE) { + /* + * Unset io-u->file to tell get_next_verify() + * that this IO is not requeue. + */ + io_u->file = NULL; + if (!get_next_verify(td, io_u)) { + zone_unlock(zb); + return io_u_accept; + } + io_u->file = f; + } + /* * Since previous write requests may have been submitted * asynchronously and since we will submit the zone @@ -1435,24 +2109,34 @@ enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) */ io_u_quiesce(td); zb->reset_zone = 0; - if (zbd_reset_zone(td, f, zb) < 0) + if (__zbd_reset_zone(td, f, zb) < 0) + goto eof; + + if (zb->capacity < min_bs) { + td_verror(td, EINVAL, "ZCAP is less min_bs"); + log_err("zone capacity %"PRIu64" smaller than minimum block size %"PRIu64"\n", + zb->capacity, min_bs); goto eof; + } } + /* Make writes occur at the write pointer */ assert(!zbd_zone_full(f, zb, min_bs)); io_u->offset = zb->wp; if (!is_valid_offset(f, io_u->offset)) { - dprint(FD_ZBD, "Dropped request with offset %llu\n", - io_u->offset); + td_verror(td, EINVAL, "invalid WP value"); + dprint(FD_ZBD, "%s: dropped request with offset %llu\n", + f->file_name, io_u->offset); goto eof; } + /* * Make sure that the buflen is a multiple of the minimal * block size. Give up if shrinking would make the request too * small. */ new_len = min((unsigned long long)io_u->buflen, - (zb + 1)->start - io_u->offset); + zbd_zone_capacity_end(zb) - io_u->offset); new_len = new_len / min_bs * min_bs; if (new_len == io_u->buflen) goto accept; @@ -1462,35 +2146,66 @@ enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u) orig_len, io_u->buflen); goto accept; } - log_err("Zone remainder %lld smaller than minimum block size %d\n", - ((zb + 1)->start - io_u->offset), - min_bs); + + td_verror(td, EIO, "zone remainder too small"); + log_err("zone remainder %lld smaller than min block size %"PRIu64"\n", + (zbd_zone_capacity_end(zb) - io_u->offset), min_bs); + goto eof; + case DDIR_TRIM: - /* fall-through */ + /* Check random trim targets a non-empty zone */ + if (!td_random(td) || zb->wp > zb->start) + goto accept; + + /* Find out a non-empty zone to trim */ + zone_unlock(zb); + zl = zbd_get_zone(f, f->max_zone); + zb = zbd_find_zone(td, io_u, 1, zb, zl); + if (zb) { + io_u->offset = zb->start; + dprint(FD_ZBD, "%s: found new zone(%lld) for trim\n", + f->file_name, io_u->offset); + goto accept; + } + + goto eof; + case DDIR_SYNC: + /* fall-through */ case DDIR_DATASYNC: case DDIR_SYNC_FILE_RANGE: case DDIR_WAIT: case DDIR_LAST: case DDIR_INVAL: + case DDIR_TIMEOUT: goto accept; } assert(false); accept: - assert(zb); - assert(zb->cond != BLK_ZONE_COND_OFFLINE); + assert(zb->has_wp); + assert(zb->cond != ZBD_ZONE_COND_OFFLINE); assert(!io_u->zbd_queue_io); assert(!io_u->zbd_put_io); + io_u->zbd_queue_io = zbd_queue_io; io_u->zbd_put_io = zbd_put_io; + + /* + * Since we return with the zone lock still held, + * add an annotation to let Coverity know that it + * is intentional. + */ + /* coverity[missing_unlock] */ + return io_u_accept; eof: - if (zb) - pthread_mutex_unlock(&zb->mutex); + if (zb && zb->has_wp) + zone_unlock(zb); + return io_u_eof; } @@ -1499,7 +2214,52 @@ char *zbd_write_status(const struct thread_stat *ts) { char *res; - if (asprintf(&res, "; %llu zone resets", (unsigned long long) ts->nr_zone_resets) < 0) + if (asprintf(&res, "; %"PRIu64" zone resets", ts->nr_zone_resets) < 0) return NULL; return res; } + +/** + * zbd_do_io_u_trim - If reset zone is applicable, do reset zone instead of trim + * + * @td: FIO thread data. + * @io_u: FIO I/O unit. + * + * It is assumed that z->mutex is already locked. + * Return io_u_completed when reset zone succeeds. Return 0 when the target zone + * does not have write pointer. On error, return negative errno. + */ +int zbd_do_io_u_trim(struct thread_data *td, struct io_u *io_u) +{ + struct fio_file *f = io_u->file; + struct fio_zone_info *z; + int ret; + + z = zbd_offset_to_zone(f, io_u->offset); + if (!z->has_wp) + return 0; + + if (io_u->offset != z->start) { + log_err("Trim offset not at zone start (%lld)\n", + io_u->offset); + return -EINVAL; + } + + ret = zbd_reset_zone((struct thread_data *)td, f, z); + if (ret < 0) + return ret; + + return io_u_completed; +} + +void zbd_log_err(const struct thread_data *td, const struct io_u *io_u) +{ + const struct fio_file *f = io_u->file; + + if (td->o.zone_mode != ZONE_MODE_ZBD) + return; + + if (io_u->error == EOVERFLOW) + log_err("%s: Exceeded max_active_zones limit. Check conditions of zones out of I/O ranges.\n", + f->file_name); +}