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 nullb_device *dev = nullb->dev;
  78         struct request_queue *q = nullb->q;
  79
  80         if (queue_is_mq(q)) {
  81                 int ret = blk_revalidate_disk_zones(nullb->disk, NULL);
  82
  83                 if (ret)
  84                         return ret;
  85         } else {
  86                 blk_queue_chunk_sectors(q, dev->zone_size_sects);
  87                 q->nr_zones = blkdev_nr_zones(nullb->disk);
  88         }
  89
  90         blk_queue_max_zone_append_sectors(q, dev->zone_size_sects);
  91
  92         return 0;
  93 }
  94
  95 void null_free_zoned_dev(struct nullb_device *dev)
  96 {
  97         kvfree(dev->zones);
  98 }
  99
 100 int null_report_zones(struct gendisk *disk, sector_t sector,
 101                 unsigned int nr_zones, report_zones_cb cb, void *data)
 102 {
 103         struct nullb *nullb = disk->private_data;
 104         struct nullb_device *dev = nullb->dev;
 105         unsigned int first_zone, i;
 106         struct blk_zone zone;
 107         int error;
 108
 109         first_zone = null_zone_no(dev, sector);
 110         if (first_zone >= dev->nr_zones)
 111                 return 0;
 112
 113         nr_zones = min(nr_zones, dev->nr_zones - first_zone);
 114         trace_nullb_report_zones(nullb, nr_zones);
 115
 116         for (i = 0; i < nr_zones; i++) {
 117                 /*
 118                  * Stacked DM target drivers will remap the zone information by
 119                  * modifying the zone information passed to the report callback.
 120                  * So use a local copy to avoid corruption of the device zone
 121                  * array.
 122                  */
 123                 memcpy(&zone, &dev->zones[first_zone + i],
 124                        sizeof(struct blk_zone));
 125                 error = cb(&zone, i, data);
 126                 if (error)
 127                         return error;
 128         }
 129
 130         return nr_zones;
 131 }
 132
 133 size_t null_zone_valid_read_len(struct nullb *nullb,
 134                                 sector_t sector, unsigned int len)
 135 {
 136         struct nullb_device *dev = nullb->dev;
 137         struct blk_zone *zone = &dev->zones[null_zone_no(dev, sector)];
 138         unsigned int nr_sectors = len >> SECTOR_SHIFT;
 139
 140         /* Read must be below the write pointer position */
 141         if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL ||
 142             sector + nr_sectors <= zone->wp)
 143                 return len;
 144
 145         if (sector > zone->wp)
 146                 return 0;
 147
 148         return (zone->wp - sector) << SECTOR_SHIFT;
 149 }
 150
 151 static blk_status_t null_zone_write(struct nullb_cmd *cmd, sector_t sector,
 152                                     unsigned int nr_sectors, bool append)
 153 {
 154         struct nullb_device *dev = cmd->nq->dev;
 155         unsigned int zno = null_zone_no(dev, sector);
 156         struct blk_zone *zone = &dev->zones[zno];
 157         blk_status_t ret;
 158
 159         trace_nullb_zone_op(cmd, zno, zone->cond);
 160
 161         if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 162                 return null_process_cmd(cmd, REQ_OP_WRITE, sector, nr_sectors);
 163
 164         switch (zone->cond) {
 165         case BLK_ZONE_COND_FULL:
 166                 /* Cannot write to a full zone */
 167                 return BLK_STS_IOERR;
 168         case BLK_ZONE_COND_EMPTY:
 169         case BLK_ZONE_COND_IMP_OPEN:
 170         case BLK_ZONE_COND_EXP_OPEN:
 171         case BLK_ZONE_COND_CLOSED:
 172                 /*
 173                  * Regular writes must be at the write pointer position.
 174                  * Zone append writes are automatically issued at the write
 175                  * pointer and the position returned using the request or BIO
 176                  * sector.
 177                  */
 178                 if (append) {
 179                         sector = zone->wp;
 180                         if (cmd->bio)
 181                                 cmd->bio->bi_iter.bi_sector = sector;
 182                         else
 183                                 cmd->rq->__sector = sector;
 184                 } else if (sector != zone->wp) {
 185                         return BLK_STS_IOERR;
 186                 }
 187
 188                 if (zone->cond != BLK_ZONE_COND_EXP_OPEN)
 189                         zone->cond = BLK_ZONE_COND_IMP_OPEN;
 190
 191                 ret = null_process_cmd(cmd, REQ_OP_WRITE, sector, nr_sectors);
 192                 if (ret != BLK_STS_OK)
 193                         return ret;
 194
 195                 zone->wp += nr_sectors;
 196                 if (zone->wp == zone->start + zone->len)
 197                         zone->cond = BLK_ZONE_COND_FULL;
 198                 return BLK_STS_OK;
 199         default:
 200                 /* Invalid zone condition */
 201                 return BLK_STS_IOERR;
 202         }
 203 }
 204
 205 static blk_status_t null_zone_mgmt(struct nullb_cmd *cmd, enum req_opf op,
 206                                    sector_t sector)
 207 {
 208         struct nullb_device *dev = cmd->nq->dev;
 209         unsigned int zone_no = null_zone_no(dev, sector);
 210         struct blk_zone *zone = &dev->zones[zone_no];
 211         size_t i;
 212
 213         switch (op) {
 214         case REQ_OP_ZONE_RESET_ALL:
 215                 for (i = 0; i < dev->nr_zones; i++) {
 216                         if (zone[i].type == BLK_ZONE_TYPE_CONVENTIONAL)
 217                                 continue;
 218                         zone[i].cond = BLK_ZONE_COND_EMPTY;
 219                         zone[i].wp = zone[i].start;
 220                 }
 221                 break;
 222         case REQ_OP_ZONE_RESET:
 223                 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 224                         return BLK_STS_IOERR;
 225
 226                 zone->cond = BLK_ZONE_COND_EMPTY;
 227                 zone->wp = zone->start;
 228                 break;
 229         case REQ_OP_ZONE_OPEN:
 230                 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 231                         return BLK_STS_IOERR;
 232                 if (zone->cond == BLK_ZONE_COND_FULL)
 233                         return BLK_STS_IOERR;
 234
 235                 zone->cond = BLK_ZONE_COND_EXP_OPEN;
 236                 break;
 237         case REQ_OP_ZONE_CLOSE:
 238                 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 239                         return BLK_STS_IOERR;
 240                 if (zone->cond == BLK_ZONE_COND_FULL)
 241                         return BLK_STS_IOERR;
 242
 243                 if (zone->wp == zone->start)
 244                         zone->cond = BLK_ZONE_COND_EMPTY;
 245                 else
 246                         zone->cond = BLK_ZONE_COND_CLOSED;
 247                 break;
 248         case REQ_OP_ZONE_FINISH:
 249                 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
 250                         return BLK_STS_IOERR;
 251
 252                 zone->cond = BLK_ZONE_COND_FULL;
 253                 zone->wp = zone->start + zone->len;
 254                 break;
 255         default:
 256                 return BLK_STS_NOTSUPP;
 257         }
 258
 259         trace_nullb_zone_op(cmd, zone_no, zone->cond);
 260         return BLK_STS_OK;
 261 }
 262
 263 blk_status_t null_process_zoned_cmd(struct nullb_cmd *cmd, enum req_opf op,
 264                                     sector_t sector, sector_t nr_sectors)
 265 {
 266         switch (op) {
 267         case REQ_OP_WRITE:
 268                 return null_zone_write(cmd, sector, nr_sectors, false);
 269         case REQ_OP_ZONE_APPEND:
 270                 return null_zone_write(cmd, sector, nr_sectors, true);
 271         case REQ_OP_ZONE_RESET:
 272         case REQ_OP_ZONE_RESET_ALL:
 273         case REQ_OP_ZONE_OPEN:
 274         case REQ_OP_ZONE_CLOSE:
 275         case REQ_OP_ZONE_FINISH:
 276                 return null_zone_mgmt(cmd, op, sector);
 277         default:
 278                 return null_process_cmd(cmd, op, sector, nr_sectors);
 279         }
 280 }