drivers/block/null_blk_zoned.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/vmalloc.h>
   3 #include "null_blk.h"
   4
   5 #define CREATE_TRACE_POINTS
   6 #include "null_blk_trace.h"
   7
   8 /* zone_size in MBs to sectors. */
   9 #define ZONE_SIZE_SHIFT         11
  10
  11 static inline unsigned int null_zone_no(struct nullb_device *dev, sector_t sect)
  12 {
  13         return sect >> ilog2(dev->zone_size_sects);
  14 }
  15
  16 int null_init_zoned_dev(struct nullb_device *dev, struct request_queue *q)
  17 {
  18         sector_t dev_size = (sector_t)dev->size * 1024 * 1024;
  19         sector_t sector = 0;
  20         unsigned int i;
  21
  22         if (!is_power_of_2(dev->zone_size)) {
  23                 pr_err("zone_size must be power-of-two\n");
  24                 return -EINVAL;
  25         }
  26         if (dev->zone_size > dev->size) {
  27                 pr_err("Zone size larger than device capacity\n");
  28                 return -EINVAL;
  29         }
  30
  31         dev->zone_size_sects = dev->zone_size << ZONE_SIZE_SHIFT;
  32         dev->nr_zones = dev_size >>
  33                                 (SECTOR_SHIFT + ilog2(dev->zone_size_sects));
  34         dev->zones = kvmalloc_array(dev->nr_zones, sizeof(struct blk_zone),
  35                         GFP_KERNEL | __GFP_ZERO);
  36         if (!dev->zones)
  37                 return -ENOMEM;
  38
  39         if (dev->zone_nr_conv >= dev->nr_zones) {
  40                 dev->zone_nr_conv = dev->nr_zones - 1;
  41                 pr_info("changed the number of conventional zones to %u",
  42                         dev->zone_nr_conv);
  43         }
  44
  45         for (i = 0; i <  dev->zone_nr_conv; i++) {
  46                 struct blk_zone *zone = &dev->zones[i];
  47
  48                 zone->start = sector;
  49                 zone->len = dev->zone_size_sects;
  50                 zone->wp = zone->start + zone->len;
  51                 zone->type = BLK_ZONE_TYPE_CONVENTIONAL;
  52                 zone->cond = BLK_ZONE_COND_NOT_WP;
  53
  54                 sector += dev->zone_size_sects;
  55         }
  56
  57         for (i = dev->zone_nr_conv; i < dev->nr_zones; i++) {
  58                 struct blk_zone *zone = &dev->zones[i];
  59
  60                 zone->start = zone->wp = sector;
  61                 zone->len = dev->zone_size_sects;
  62                 zone->type = BLK_ZONE_TYPE_SEQWRITE_REQ;
  63                 zone->cond = BLK_ZONE_COND_EMPTY;
  64
  65                 sector += dev->zone_size_sects;
  66         }
  67
  68         q->limits.zoned = BLK_ZONED_HM;
  69         blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
  70         blk_queue_required_elevator_features(q, ELEVATOR_F_ZBD_SEQ_WRITE);
  71
  72         return 0;
  73 }
  74
  75 int null_register_zoned_dev(struct nullb *nullb)
  76 {
  77         struct request_queue *q = nullb->q;
  78
  79         if (queue_is_mq(q))
  80                 return blk_revalidate_disk_zones(nullb->disk);
  81
  82         blk_queue_chunk_sectors(q, nullb->dev->zone_size_sects);
  83         q->nr_zones = blkdev_nr_zones(nullb->disk);
  84
  85         return 0;
  86 }
  87
  88 void null_free_zoned_dev(struct nullb_device *dev)
  89 {
  90         kvfree(dev->zones);
  91 }
  92
  93 int null_report_zones(struct gendisk *disk, sector_t sector,
  94                 unsigned int nr_zones, report_zones_cb cb, void *data)
  95 {
  96         struct nullb *nullb = disk->private_data;
  97         struct nullb_device *dev = nullb->dev;
  98         unsigned int first_zone, i;
  99         struct blk_zone zone;
 100         int error;
 101
 102         first_zone = null_zone_no(dev, sector);
 103         if (first_zone >= dev->nr_zones)
 104                 return 0;
 105
 106         nr_zones = min(nr_zones, dev->nr_zones - first_zone);
 107         trace_nullb_report_zones(nullb, nr_zones);
 108
 109         for (i = 0; i < nr_zones; i++) {
 110                 /*
 111                  * Stacked DM target drivers will remap the zone information by
 112                  * modifying the zone information passed to the report callback.
 113                  * So use a local copy to avoid corruption of the device zone
 114                  * array.
 115                  */
 116                 memcpy(&zone, &dev->zones[first_zone + i],
 117                        sizeof(struct blk_zone));
 118                 error = cb(&zone, i, data);
 119                 if (error)
 120                         return error;
 121         }
 122
 123         return nr_zones;
 124 }
 125
 126 size_t null_zone_valid_read_len(struct nullb *nullb,
 127                                 sector_t sector, unsigned int len)
 128 {
 129         struct nullb_device *dev = nullb->dev;
 130         struct blk_zone *zone = &dev->zones[null_zone_no(dev, sector)];
 131         unsigned int nr_sectors = len >> SECTOR_SHIFT;
 132
 133         /* Read must be below the write pointer position */
 134         if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL ||
 135             sector + nr_sectors <= zone->wp)
 136                 return len;
 137
 138         if (sector > zone->wp)
 139                 return 0;
 140
 141         return (zone->wp - sector) << SECTOR_SHIFT;
 142 }
 143
 144 static blk_status_t null_zone_write(struct nullb_cmd *cmd, sector_t sector,
 145                      unsigned int nr_sectors)
 146 {
 147         struct nullb_device *dev = cmd->nq->dev;
 148         unsigned int zno = null_zone_no(dev, sector);
 149         struct blk_zone *zone = &dev->zones[zno];
 150         blk_status_t ret;
 151
 152         trace_nullb_zone_op(cmd, zno, zone->cond);
 153
 154         if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 155                 return null_process_cmd(cmd, REQ_OP_WRITE, sector, nr_sectors);
 156
 157         switch (zone->cond) {
 158         case BLK_ZONE_COND_FULL:
 159                 /* Cannot write to a full zone */
 160                 return BLK_STS_IOERR;
 161         case BLK_ZONE_COND_EMPTY:
 162         case BLK_ZONE_COND_IMP_OPEN:
 163         case BLK_ZONE_COND_EXP_OPEN:
 164         case BLK_ZONE_COND_CLOSED:
 165                 /* Writes must be at the write pointer position */
 166                 if (sector != zone->wp)
 167                         return BLK_STS_IOERR;
 168
 169                 if (zone->cond != BLK_ZONE_COND_EXP_OPEN)
 170                         zone->cond = BLK_ZONE_COND_IMP_OPEN;
 171
 172                 ret = null_process_cmd(cmd, REQ_OP_WRITE, sector, nr_sectors);
 173                 if (ret != BLK_STS_OK)
 174                         return ret;
 175
 176                 zone->wp += nr_sectors;
 177                 if (zone->wp == zone->start + zone->len)
 178                         zone->cond = BLK_ZONE_COND_FULL;
 179                 return BLK_STS_OK;
 180         default:
 181                 /* Invalid zone condition */
 182                 return BLK_STS_IOERR;
 183         }
 184 }
 185
 186 static blk_status_t null_zone_mgmt(struct nullb_cmd *cmd, enum req_opf op,
 187                                    sector_t sector)
 188 {
 189         struct nullb_device *dev = cmd->nq->dev;
 190         unsigned int zone_no = null_zone_no(dev, sector);
 191         struct blk_zone *zone = &dev->zones[zone_no];
 192         size_t i;
 193
 194         switch (op) {
 195         case REQ_OP_ZONE_RESET_ALL:
 196                 for (i = 0; i < dev->nr_zones; i++) {
 197                         if (zone[i].type == BLK_ZONE_TYPE_CONVENTIONAL)
 198                                 continue;
 199                         zone[i].cond = BLK_ZONE_COND_EMPTY;
 200                         zone[i].wp = zone[i].start;
 201                 }
 202                 break;
 203         case REQ_OP_ZONE_RESET:
 204                 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 205                         return BLK_STS_IOERR;
 206
 207                 zone->cond = BLK_ZONE_COND_EMPTY;
 208                 zone->wp = zone->start;
 209                 break;
 210         case REQ_OP_ZONE_OPEN:
 211                 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 212                         return BLK_STS_IOERR;
 213                 if (zone->cond == BLK_ZONE_COND_FULL)
 214                         return BLK_STS_IOERR;
 215
 216                 zone->cond = BLK_ZONE_COND_EXP_OPEN;
 217                 break;
 218         case REQ_OP_ZONE_CLOSE:
 219                 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 220                         return BLK_STS_IOERR;
 221                 if (zone->cond == BLK_ZONE_COND_FULL)
 222                         return BLK_STS_IOERR;
 223
 224                 if (zone->wp == zone->start)
 225                         zone->cond = BLK_ZONE_COND_EMPTY;
 226                 else
 227                         zone->cond = BLK_ZONE_COND_CLOSED;
 228                 break;
 229         case REQ_OP_ZONE_FINISH:
 230                 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 231                         return BLK_STS_IOERR;
 232
 233                 zone->cond = BLK_ZONE_COND_FULL;
 234                 zone->wp = zone->start + zone->len;
 235                 break;
 236         default:
 237                 return BLK_STS_NOTSUPP;
 238         }
 239
 240         trace_nullb_zone_op(cmd, zone_no, zone->cond);
 241         return BLK_STS_OK;
 242 }
 243
 244 blk_status_t null_process_zoned_cmd(struct nullb_cmd *cmd, enum req_opf op,
 245                                     sector_t sector, sector_t nr_sectors)
 246 {
 247         switch (op) {
 248         case REQ_OP_WRITE:
 249                 return null_zone_write(cmd, sector, nr_sectors);
 250         case REQ_OP_ZONE_RESET:
 251         case REQ_OP_ZONE_RESET_ALL:
 252         case REQ_OP_ZONE_OPEN:
 253         case REQ_OP_ZONE_CLOSE:
 254         case REQ_OP_ZONE_FINISH:
 255                 return null_zone_mgmt(cmd, op, sector);
 256         default:
 257                 return null_process_cmd(cmd, op, sector, nr_sectors);
 258         }
 259 }