drivers/scsi/sd_zbc.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * SCSI Zoned Block commands
   4  *
   5  * Copyright (C) 2014-2015 SUSE Linux GmbH
   6  * Written by: Hannes Reinecke <hare@suse.de>
   7  * Modified by: Damien Le Moal <damien.lemoal@hgst.com>
   8  * Modified by: Shaun Tancheff <shaun.tancheff@seagate.com>
   9  */
  10
  11 #include <linux/blkdev.h>
  12 #include <linux/vmalloc.h>
  13 #include <linux/sched/mm.h>
  14 #include <linux/mutex.h>
  15
  16 #include <asm/unaligned.h>
  17
  18 #include <scsi/scsi.h>
  19 #include <scsi/scsi_cmnd.h>
  20
  21 #include "sd.h"
  22
  23 static unsigned int sd_zbc_get_zone_wp_offset(struct blk_zone *zone)
  24 {
  25         if (zone->type == ZBC_ZONE_TYPE_CONV)
  26                 return 0;
  27
  28         switch (zone->cond) {
  29         case BLK_ZONE_COND_IMP_OPEN:
  30         case BLK_ZONE_COND_EXP_OPEN:
  31         case BLK_ZONE_COND_CLOSED:
  32                 return zone->wp - zone->start;
  33         case BLK_ZONE_COND_FULL:
  34                 return zone->len;
  35         case BLK_ZONE_COND_EMPTY:
  36         case BLK_ZONE_COND_OFFLINE:
  37         case BLK_ZONE_COND_READONLY:
  38         default:
  39                 /*
  40                  * Offline and read-only zones do not have a valid
  41                  * write pointer. Use 0 as for an empty zone.
  42                  */
  43                 return 0;
  44         }
  45 }
  46
  47 static int sd_zbc_parse_report(struct scsi_disk *sdkp, u8 *buf,
  48                                unsigned int idx, report_zones_cb cb, void *data)
  49 {
  50         struct scsi_device *sdp = sdkp->device;
  51         struct blk_zone zone = { 0 };
  52         int ret;
  53
  54         zone.type = buf[0] & 0x0f;
  55         zone.cond = (buf[1] >> 4) & 0xf;
  56         if (buf[1] & 0x01)
  57                 zone.reset = 1;
  58         if (buf[1] & 0x02)
  59                 zone.non_seq = 1;
  60
  61         zone.len = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
  62         zone.start = logical_to_sectors(sdp, get_unaligned_be64(&buf[16]));
  63         zone.wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24]));
  64         if (zone.type != ZBC_ZONE_TYPE_CONV &&
  65             zone.cond == ZBC_ZONE_COND_FULL)
  66                 zone.wp = zone.start + zone.len;
  67
  68         ret = cb(&zone, idx, data);
  69         if (ret)
  70                 return ret;
  71
  72         if (sdkp->rev_wp_offset)
  73                 sdkp->rev_wp_offset[idx] = sd_zbc_get_zone_wp_offset(&zone);
  74
  75         return 0;
  76 }
  77
  78 /**
  79  * sd_zbc_do_report_zones - Issue a REPORT ZONES scsi command.
  80  * @sdkp: The target disk
  81  * @buf: vmalloc-ed buffer to use for the reply
  82  * @buflen: the buffer size
  83  * @lba: Start LBA of the report
  84  * @partial: Do partial report
  85  *
  86  * For internal use during device validation.
  87  * Using partial=true can significantly speed up execution of a report zones
  88  * command because the disk does not have to count all possible report matching
  89  * zones and will only report the count of zones fitting in the command reply
  90  * buffer.
  91  */
  92 static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
  93                                   unsigned int buflen, sector_t lba,
  94                                   bool partial)
  95 {
  96         struct scsi_device *sdp = sdkp->device;
  97         const int timeout = sdp->request_queue->rq_timeout;
  98         struct scsi_sense_hdr sshdr;
  99         unsigned char cmd[16];
 100         unsigned int rep_len;
 101         int result;
 102
 103         memset(cmd, 0, 16);
 104         cmd[0] = ZBC_IN;
 105         cmd[1] = ZI_REPORT_ZONES;
 106         put_unaligned_be64(lba, &cmd[2]);
 107         put_unaligned_be32(buflen, &cmd[10]);
 108         if (partial)
 109                 cmd[14] = ZBC_REPORT_ZONE_PARTIAL;
 110
 111         result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
 112                                   buf, buflen, &sshdr,
 113                                   timeout, SD_MAX_RETRIES, NULL);
 114         if (result) {
 115                 sd_printk(KERN_ERR, sdkp,
 116                           "REPORT ZONES start lba %llu failed\n", lba);
 117                 sd_print_result(sdkp, "REPORT ZONES", result);
 118                 if (driver_byte(result) == DRIVER_SENSE &&
 119                     scsi_sense_valid(&sshdr))
 120                         sd_print_sense_hdr(sdkp, &sshdr);
 121                 return -EIO;
 122         }
 123
 124         rep_len = get_unaligned_be32(&buf[0]);
 125         if (rep_len < 64) {
 126                 sd_printk(KERN_ERR, sdkp,
 127                           "REPORT ZONES report invalid length %u\n",
 128                           rep_len);
 129                 return -EIO;
 130         }
 131
 132         return 0;
 133 }
 134
 135 /**
 136  * Allocate a buffer for report zones reply.
 137  * @sdkp: The target disk
 138  * @nr_zones: Maximum number of zones to report
 139  * @buflen: Size of the buffer allocated
 140  *
 141  * Try to allocate a reply buffer for the number of requested zones.
 142  * The size of the buffer allocated may be smaller than requested to
 143  * satify the device constraint (max_hw_sectors, max_segments, etc).
 144  *
 145  * Return the address of the allocated buffer and update @buflen with
 146  * the size of the allocated buffer.
 147  */
 148 static void *sd_zbc_alloc_report_buffer(struct scsi_disk *sdkp,
 149                                         unsigned int nr_zones, size_t *buflen)
 150 {
 151         struct request_queue *q = sdkp->disk->queue;
 152         size_t bufsize;
 153         void *buf;
 154
 155         /*
 156          * Report zone buffer size should be at most 64B times the number of
 157          * zones requested plus the 64B reply header, but should be at least
 158          * SECTOR_SIZE for ATA devices.
 159          * Make sure that this size does not exceed the hardware capabilities.
 160          * Furthermore, since the report zone command cannot be split, make
 161          * sure that the allocated buffer can always be mapped by limiting the
 162          * number of pages allocated to the HBA max segments limit.
 163          */
 164         nr_zones = min(nr_zones, sdkp->nr_zones);
 165         bufsize = roundup((nr_zones + 1) * 64, SECTOR_SIZE);
 166         bufsize = min_t(size_t, bufsize,
 167                         queue_max_hw_sectors(q) << SECTOR_SHIFT);
 168         bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
 169
 170         while (bufsize >= SECTOR_SIZE) {
 171                 buf = __vmalloc(bufsize,
 172                                 GFP_KERNEL | __GFP_ZERO | __GFP_NORETRY);
 173                 if (buf) {
 174                         *buflen = bufsize;
 175                         return buf;
 176                 }
 177                 bufsize >>= 1;
 178         }
 179
 180         return NULL;
 181 }
 182
 183 /**
 184  * sd_zbc_zone_sectors - Get the device zone size in number of 512B sectors.
 185  * @sdkp: The target disk
 186  */
 187 static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
 188 {
 189         return logical_to_sectors(sdkp->device, sdkp->zone_blocks);
 190 }
 191
 192 int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 193                         unsigned int nr_zones, report_zones_cb cb, void *data)
 194 {
 195         struct scsi_disk *sdkp = scsi_disk(disk);
 196         sector_t capacity = logical_to_sectors(sdkp->device, sdkp->capacity);
 197         unsigned int nr, i;
 198         unsigned char *buf;
 199         size_t offset, buflen = 0;
 200         int zone_idx = 0;
 201         int ret;
 202
 203         if (!sd_is_zoned(sdkp))
 204                 /* Not a zoned device */
 205                 return -EOPNOTSUPP;
 206
 207         if (!capacity)
 208                 /* Device gone or invalid */
 209                 return -ENODEV;
 210
 211         buf = sd_zbc_alloc_report_buffer(sdkp, nr_zones, &buflen);
 212         if (!buf)
 213                 return -ENOMEM;
 214
 215         while (zone_idx < nr_zones && sector < capacity) {
 216                 ret = sd_zbc_do_report_zones(sdkp, buf, buflen,
 217                                 sectors_to_logical(sdkp->device, sector), true);
 218                 if (ret)
 219                         goto out;
 220
 221                 offset = 0;
 222                 nr = min(nr_zones, get_unaligned_be32(&buf[0]) / 64);
 223                 if (!nr)
 224                         break;
 225
 226                 for (i = 0; i < nr && zone_idx < nr_zones; i++) {
 227                         offset += 64;
 228                         ret = sd_zbc_parse_report(sdkp, buf + offset, zone_idx,
 229                                                   cb, data);
 230                         if (ret)
 231                                 goto out;
 232                         zone_idx++;
 233                 }
 234
 235                 sector += sd_zbc_zone_sectors(sdkp) * i;
 236         }
 237
 238         ret = zone_idx;
 239 out:
 240         kvfree(buf);
 241         return ret;
 242 }
 243
 244 static blk_status_t sd_zbc_cmnd_checks(struct scsi_cmnd *cmd)
 245 {
 246         struct request *rq = cmd->request;
 247         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 248         sector_t sector = blk_rq_pos(rq);
 249
 250         if (!sd_is_zoned(sdkp))
 251                 /* Not a zoned device */
 252                 return BLK_STS_IOERR;
 253
 254         if (sdkp->device->changed)
 255                 return BLK_STS_IOERR;
 256
 257         if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
 258                 /* Unaligned request */
 259                 return BLK_STS_IOERR;
 260
 261         return BLK_STS_OK;
 262 }
 263
 264 #define SD_ZBC_INVALID_WP_OFST  (~0u)
 265 #define SD_ZBC_UPDATING_WP_OFST (SD_ZBC_INVALID_WP_OFST - 1)
 266
 267 static int sd_zbc_update_wp_offset_cb(struct blk_zone *zone, unsigned int idx,
 268                                     void *data)
 269 {
 270         struct scsi_disk *sdkp = data;
 271
 272         lockdep_assert_held(&sdkp->zones_wp_offset_lock);
 273
 274         sdkp->zones_wp_offset[idx] = sd_zbc_get_zone_wp_offset(zone);
 275
 276         return 0;
 277 }
 278
 279 static void sd_zbc_update_wp_offset_workfn(struct work_struct *work)
 280 {
 281         struct scsi_disk *sdkp;
 282         unsigned int zno;
 283         int ret;
 284
 285         sdkp = container_of(work, struct scsi_disk, zone_wp_offset_work);
 286
 287         spin_lock_bh(&sdkp->zones_wp_offset_lock);
 288         for (zno = 0; zno < sdkp->nr_zones; zno++) {
 289                 if (sdkp->zones_wp_offset[zno] != SD_ZBC_UPDATING_WP_OFST)
 290                         continue;
 291
 292                 spin_unlock_bh(&sdkp->zones_wp_offset_lock);
 293                 ret = sd_zbc_do_report_zones(sdkp, sdkp->zone_wp_update_buf,
 294                                              SD_BUF_SIZE,
 295                                              zno * sdkp->zone_blocks, true);
 296                 spin_lock_bh(&sdkp->zones_wp_offset_lock);
 297                 if (!ret)
 298                         sd_zbc_parse_report(sdkp, sdkp->zone_wp_update_buf + 64,
 299                                             zno, sd_zbc_update_wp_offset_cb,
 300                                             sdkp);
 301         }
 302         spin_unlock_bh(&sdkp->zones_wp_offset_lock);
 303
 304         scsi_device_put(sdkp->device);
 305 }
 306
 307 /**
 308  * sd_zbc_prepare_zone_append() - Prepare an emulated ZONE_APPEND command.
 309  * @cmd: the command to setup
 310  * @lba: the LBA to patch
 311  * @nr_blocks: the number of LBAs to be written
 312  *
 313  * Called from sd_setup_read_write_cmnd() for REQ_OP_ZONE_APPEND.
 314  * @sd_zbc_prepare_zone_append() handles the necessary zone wrote locking and
 315  * patching of the lba for an emulated ZONE_APPEND command.
 316  *
 317  * In case the cached write pointer offset is %SD_ZBC_INVALID_WP_OFST it will
 318  * schedule a REPORT ZONES command and return BLK_STS_IOERR.
 319  */
 320 blk_status_t sd_zbc_prepare_zone_append(struct scsi_cmnd *cmd, sector_t *lba,
 321                                         unsigned int nr_blocks)
 322 {
 323         struct request *rq = cmd->request;
 324         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 325         unsigned int wp_offset, zno = blk_rq_zone_no(rq);
 326         blk_status_t ret;
 327
 328         ret = sd_zbc_cmnd_checks(cmd);
 329         if (ret != BLK_STS_OK)
 330                 return ret;
 331
 332         if (!blk_rq_zone_is_seq(rq))
 333                 return BLK_STS_IOERR;
 334
 335         /* Unlock of the write lock will happen in sd_zbc_complete() */
 336         if (!blk_req_zone_write_trylock(rq))
 337                 return BLK_STS_ZONE_RESOURCE;
 338
 339         spin_lock_bh(&sdkp->zones_wp_offset_lock);
 340         wp_offset = sdkp->zones_wp_offset[zno];
 341         switch (wp_offset) {
 342         case SD_ZBC_INVALID_WP_OFST:
 343                 /*
 344                  * We are about to schedule work to update a zone write pointer
 345                  * offset, which will cause the zone append command to be
 346                  * requeued. So make sure that the scsi device does not go away
 347                  * while the work is being processed.
 348                  */
 349                 if (scsi_device_get(sdkp->device)) {
 350                         ret = BLK_STS_IOERR;
 351                         break;
 352                 }
 353                 sdkp->zones_wp_offset[zno] = SD_ZBC_UPDATING_WP_OFST;
 354                 schedule_work(&sdkp->zone_wp_offset_work);
 355                 fallthrough;
 356         case SD_ZBC_UPDATING_WP_OFST:
 357                 ret = BLK_STS_DEV_RESOURCE;
 358                 break;
 359         default:
 360                 wp_offset = sectors_to_logical(sdkp->device, wp_offset);
 361                 if (wp_offset + nr_blocks > sdkp->zone_blocks) {
 362                         ret = BLK_STS_IOERR;
 363                         break;
 364                 }
 365
 366                 *lba += wp_offset;
 367         }
 368         spin_unlock_bh(&sdkp->zones_wp_offset_lock);
 369         if (ret)
 370                 blk_req_zone_write_unlock(rq);
 371         return ret;
 372 }
 373
 374 /**
 375  * sd_zbc_setup_zone_mgmt_cmnd - Prepare a zone ZBC_OUT command. The operations
 376  *                      can be RESET WRITE POINTER, OPEN, CLOSE or FINISH.
 377  * @cmd: the command to setup
 378  * @op: Operation to be performed
 379  * @all: All zones control
 380  *
 381  * Called from sd_init_command() for REQ_OP_ZONE_RESET, REQ_OP_ZONE_RESET_ALL,
 382  * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE or REQ_OP_ZONE_FINISH requests.
 383  */
 384 blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
 385                                          unsigned char op, bool all)
 386 {
 387         struct request *rq = cmd->request;
 388         sector_t sector = blk_rq_pos(rq);
 389         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 390         sector_t block = sectors_to_logical(sdkp->device, sector);
 391         blk_status_t ret;
 392
 393         ret = sd_zbc_cmnd_checks(cmd);
 394         if (ret != BLK_STS_OK)
 395                 return ret;
 396
 397         cmd->cmd_len = 16;
 398         memset(cmd->cmnd, 0, cmd->cmd_len);
 399         cmd->cmnd[0] = ZBC_OUT;
 400         cmd->cmnd[1] = op;
 401         if (all)
 402                 cmd->cmnd[14] = 0x1;
 403         else
 404                 put_unaligned_be64(block, &cmd->cmnd[2]);
 405
 406         rq->timeout = SD_TIMEOUT;
 407         cmd->sc_data_direction = DMA_NONE;
 408         cmd->transfersize = 0;
 409         cmd->allowed = 0;
 410
 411         return BLK_STS_OK;
 412 }
 413
 414 static bool sd_zbc_need_zone_wp_update(struct request *rq)
 415 {
 416         switch (req_op(rq)) {
 417         case REQ_OP_ZONE_APPEND:
 418         case REQ_OP_ZONE_FINISH:
 419         case REQ_OP_ZONE_RESET:
 420         case REQ_OP_ZONE_RESET_ALL:
 421                 return true;
 422         case REQ_OP_WRITE:
 423         case REQ_OP_WRITE_ZEROES:
 424         case REQ_OP_WRITE_SAME:
 425                 return blk_rq_zone_is_seq(rq);
 426         default:
 427                 return false;
 428         }
 429 }
 430
 431 /**
 432  * sd_zbc_zone_wp_update - Update cached zone write pointer upon cmd completion
 433  * @cmd: Completed command
 434  * @good_bytes: Command reply bytes
 435  *
 436  * Called from sd_zbc_complete() to handle the update of the cached zone write
 437  * pointer value in case an update is needed.
 438  */
 439 static unsigned int sd_zbc_zone_wp_update(struct scsi_cmnd *cmd,
 440                                           unsigned int good_bytes)
 441 {
 442         int result = cmd->result;
 443         struct request *rq = cmd->request;
 444         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 445         unsigned int zno = blk_rq_zone_no(rq);
 446         enum req_opf op = req_op(rq);
 447
 448         /*
 449          * If we got an error for a command that needs updating the write
 450          * pointer offset cache, we must mark the zone wp offset entry as
 451          * invalid to force an update from disk the next time a zone append
 452          * command is issued.
 453          */
 454         spin_lock_bh(&sdkp->zones_wp_offset_lock);
 455
 456         if (result && op != REQ_OP_ZONE_RESET_ALL) {
 457                 if (op == REQ_OP_ZONE_APPEND) {
 458                         /* Force complete completion (no retry) */
 459                         good_bytes = 0;
 460                         scsi_set_resid(cmd, blk_rq_bytes(rq));
 461                 }
 462
 463                 /*
 464                  * Force an update of the zone write pointer offset on
 465                  * the next zone append access.
 466                  */
 467                 if (sdkp->zones_wp_offset[zno] != SD_ZBC_UPDATING_WP_OFST)
 468                         sdkp->zones_wp_offset[zno] = SD_ZBC_INVALID_WP_OFST;
 469                 goto unlock_wp_offset;
 470         }
 471
 472         switch (op) {
 473         case REQ_OP_ZONE_APPEND:
 474                 rq->__sector += sdkp->zones_wp_offset[zno];
 475                 fallthrough;
 476         case REQ_OP_WRITE_ZEROES:
 477         case REQ_OP_WRITE_SAME:
 478         case REQ_OP_WRITE:
 479                 if (sdkp->zones_wp_offset[zno] < sd_zbc_zone_sectors(sdkp))
 480                         sdkp->zones_wp_offset[zno] +=
 481                                                 good_bytes >> SECTOR_SHIFT;
 482                 break;
 483         case REQ_OP_ZONE_RESET:
 484                 sdkp->zones_wp_offset[zno] = 0;
 485                 break;
 486         case REQ_OP_ZONE_FINISH:
 487                 sdkp->zones_wp_offset[zno] = sd_zbc_zone_sectors(sdkp);
 488                 break;
 489         case REQ_OP_ZONE_RESET_ALL:
 490                 memset(sdkp->zones_wp_offset, 0,
 491                        sdkp->nr_zones * sizeof(unsigned int));
 492                 break;
 493         default:
 494                 break;
 495         }
 496
 497 unlock_wp_offset:
 498         spin_unlock_bh(&sdkp->zones_wp_offset_lock);
 499
 500         return good_bytes;
 501 }
 502
 503 /**
 504  * sd_zbc_complete - ZBC command post processing.
 505  * @cmd: Completed command
 506  * @good_bytes: Command reply bytes
 507  * @sshdr: command sense header
 508  *
 509  * Called from sd_done() to handle zone commands errors and updates to the
 510  * device queue zone write pointer offset cahce.
 511  */
 512 unsigned int sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
 513                      struct scsi_sense_hdr *sshdr)
 514 {
 515         int result = cmd->result;
 516         struct request *rq = cmd->request;
 517
 518         if (op_is_zone_mgmt(req_op(rq)) &&
 519             result &&
 520             sshdr->sense_key == ILLEGAL_REQUEST &&
 521             sshdr->asc == 0x24) {
 522                 /*
 523                  * INVALID FIELD IN CDB error: a zone management command was
 524                  * attempted on a conventional zone. Nothing to worry about,
 525                  * so be quiet about the error.
 526                  */
 527                 rq->rq_flags |= RQF_QUIET;
 528         } else if (sd_zbc_need_zone_wp_update(rq))
 529                 good_bytes = sd_zbc_zone_wp_update(cmd, good_bytes);
 530
 531         if (req_op(rq) == REQ_OP_ZONE_APPEND)
 532                 blk_req_zone_write_unlock(rq);
 533
 534         return good_bytes;
 535 }
 536
 537 /**
 538  * sd_zbc_check_zoned_characteristics - Check zoned block device characteristics
 539  * @sdkp: Target disk
 540  * @buf: Buffer where to store the VPD page data
 541  *
 542  * Read VPD page B6, get information and check that reads are unconstrained.
 543  */
 544 static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
 545                                               unsigned char *buf)
 546 {
 547
 548         if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
 549                 sd_printk(KERN_NOTICE, sdkp,
 550                           "Read zoned characteristics VPD page failed\n");
 551                 return -ENODEV;
 552         }
 553
 554         if (sdkp->device->type != TYPE_ZBC) {
 555                 /* Host-aware */
 556                 sdkp->urswrz = 1;
 557                 sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]);
 558                 sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]);
 559                 sdkp->zones_max_open = 0;
 560         } else {
 561                 /* Host-managed */
 562                 sdkp->urswrz = buf[4] & 1;
 563                 sdkp->zones_optimal_open = 0;
 564                 sdkp->zones_optimal_nonseq = 0;
 565                 sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
 566         }
 567
 568         /*
 569          * Check for unconstrained reads: host-managed devices with
 570          * constrained reads (drives failing read after write pointer)
 571          * are not supported.
 572          */
 573         if (!sdkp->urswrz) {
 574                 if (sdkp->first_scan)
 575                         sd_printk(KERN_NOTICE, sdkp,
 576                           "constrained reads devices are not supported\n");
 577                 return -ENODEV;
 578         }
 579
 580         return 0;
 581 }
 582
 583 /**
 584  * sd_zbc_check_capacity - Check the device capacity
 585  * @sdkp: Target disk
 586  * @buf: command buffer
 587  * @zblock: zone size in number of blocks
 588  *
 589  * Get the device zone size and check that the device capacity as reported
 590  * by READ CAPACITY matches the max_lba value (plus one) of the report zones
 591  * command reply for devices with RC_BASIS == 0.
 592  *
 593  * Returns 0 upon success or an error code upon failure.
 594  */
 595 static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf,
 596                                  u32 *zblocks)
 597 {
 598         u64 zone_blocks;
 599         sector_t max_lba;
 600         unsigned char *rec;
 601         int ret;
 602
 603         /* Do a report zone to get max_lba and the size of the first zone */
 604         ret = sd_zbc_do_report_zones(sdkp, buf, SD_BUF_SIZE, 0, false);
 605         if (ret)
 606                 return ret;
 607
 608         if (sdkp->rc_basis == 0) {
 609                 /* The max_lba field is the capacity of this device */
 610                 max_lba = get_unaligned_be64(&buf[8]);
 611                 if (sdkp->capacity != max_lba + 1) {
 612                         if (sdkp->first_scan)
 613                                 sd_printk(KERN_WARNING, sdkp,
 614                                         "Changing capacity from %llu to max LBA+1 %llu\n",
 615                                         (unsigned long long)sdkp->capacity,
 616                                         (unsigned long long)max_lba + 1);
 617                         sdkp->capacity = max_lba + 1;
 618                 }
 619         }
 620
 621         /* Get the size of the first reported zone */
 622         rec = buf + 64;
 623         zone_blocks = get_unaligned_be64(&rec[8]);
 624         if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
 625                 if (sdkp->first_scan)
 626                         sd_printk(KERN_NOTICE, sdkp,
 627                                   "Zone size too large\n");
 628                 return -EFBIG;
 629         }
 630
 631         *zblocks = zone_blocks;
 632
 633         return 0;
 634 }
 635
 636 static void sd_zbc_revalidate_zones_cb(struct gendisk *disk)
 637 {
 638         struct scsi_disk *sdkp = scsi_disk(disk);
 639
 640         swap(sdkp->zones_wp_offset, sdkp->rev_wp_offset);
 641 }
 642
 643 static int sd_zbc_revalidate_zones(struct scsi_disk *sdkp,
 644                                    u32 zone_blocks,
 645                                    unsigned int nr_zones)
 646 {
 647         struct gendisk *disk = sdkp->disk;
 648         int ret = 0;
 649
 650         /*
 651          * Make sure revalidate zones are serialized to ensure exclusive
 652          * updates of the scsi disk data.
 653          */
 654         mutex_lock(&sdkp->rev_mutex);
 655
 656         /*
 657          * Revalidate the disk zones to update the device request queue zone
 658          * bitmaps and the zone write pointer offset array. Do this only once
 659          * the device capacity is set on the second revalidate execution for
 660          * disk scan or if something changed when executing a normal revalidate.
 661          */
 662         if (sdkp->first_scan) {
 663                 sdkp->zone_blocks = zone_blocks;
 664                 sdkp->nr_zones = nr_zones;
 665                 goto unlock;
 666         }
 667
 668         if (sdkp->zone_blocks == zone_blocks &&
 669             sdkp->nr_zones == nr_zones &&
 670             disk->queue->nr_zones == nr_zones)
 671                 goto unlock;
 672
 673         sdkp->rev_wp_offset = kvcalloc(nr_zones, sizeof(u32), GFP_NOIO);
 674         if (!sdkp->rev_wp_offset) {
 675                 ret = -ENOMEM;
 676                 goto unlock;
 677         }
 678
 679         ret = blk_revalidate_disk_zones(disk, sd_zbc_revalidate_zones_cb);
 680
 681         kvfree(sdkp->rev_wp_offset);
 682         sdkp->rev_wp_offset = NULL;
 683
 684 unlock:
 685         mutex_unlock(&sdkp->rev_mutex);
 686
 687         return ret;
 688 }
 689
 690 int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
 691 {
 692         struct gendisk *disk = sdkp->disk;
 693         struct request_queue *q = disk->queue;
 694         unsigned int nr_zones;
 695         u32 zone_blocks = 0;
 696         u32 max_append;
 697         int ret;
 698
 699         if (!sd_is_zoned(sdkp))
 700                 /*
 701                  * Device managed or normal SCSI disk,
 702                  * no special handling required
 703                  */
 704                 return 0;
 705
 706         /* Check zoned block device characteristics (unconstrained reads) */
 707         ret = sd_zbc_check_zoned_characteristics(sdkp, buf);
 708         if (ret)
 709                 goto err;
 710
 711         /* Check the device capacity reported by report zones */
 712         ret = sd_zbc_check_capacity(sdkp, buf, &zone_blocks);
 713         if (ret != 0)
 714                 goto err;
 715
 716         /* The drive satisfies the kernel restrictions: set it up */
 717         blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
 718         blk_queue_required_elevator_features(q, ELEVATOR_F_ZBD_SEQ_WRITE);
 719         nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
 720
 721         /* READ16/WRITE16 is mandatory for ZBC disks */
 722         sdkp->device->use_16_for_rw = 1;
 723         sdkp->device->use_10_for_rw = 0;
 724
 725         ret = sd_zbc_revalidate_zones(sdkp, zone_blocks, nr_zones);
 726         if (ret)
 727                 goto err;
 728
 729         /*
 730          * On the first scan 'chunk_sectors' isn't setup yet, so calling
 731          * blk_queue_max_zone_append_sectors() will result in a WARN(). Defer
 732          * this setting to the second scan.
 733          */
 734         if (sdkp->first_scan)
 735                 return 0;
 736
 737         max_append = min_t(u32, logical_to_sectors(sdkp->device, zone_blocks),
 738                            q->limits.max_segments << (PAGE_SHIFT - 9));
 739         max_append = min_t(u32, max_append, queue_max_hw_sectors(q));
 740
 741         blk_queue_max_zone_append_sectors(q, max_append);
 742
 743         return 0;
 744
 745 err:
 746         sdkp->capacity = 0;
 747
 748         return ret;
 749 }
 750
 751 void sd_zbc_print_zones(struct scsi_disk *sdkp)
 752 {
 753         if (!sd_is_zoned(sdkp) || !sdkp->capacity)
 754                 return;
 755
 756         if (sdkp->capacity & (sdkp->zone_blocks - 1))
 757                 sd_printk(KERN_NOTICE, sdkp,
 758                           "%u zones of %u logical blocks + 1 runt zone\n",
 759                           sdkp->nr_zones - 1,
 760                           sdkp->zone_blocks);
 761         else
 762                 sd_printk(KERN_NOTICE, sdkp,
 763                           "%u zones of %u logical blocks\n",
 764                           sdkp->nr_zones,
 765                           sdkp->zone_blocks);
 766 }
 767
 768 int sd_zbc_init_disk(struct scsi_disk *sdkp)
 769 {
 770         if (!sd_is_zoned(sdkp))
 771                 return 0;
 772
 773         sdkp->zones_wp_offset = NULL;
 774         spin_lock_init(&sdkp->zones_wp_offset_lock);
 775         sdkp->rev_wp_offset = NULL;
 776         mutex_init(&sdkp->rev_mutex);
 777         INIT_WORK(&sdkp->zone_wp_offset_work, sd_zbc_update_wp_offset_workfn);
 778         sdkp->zone_wp_update_buf = kzalloc(SD_BUF_SIZE, GFP_KERNEL);
 779         if (!sdkp->zone_wp_update_buf)
 780                 return -ENOMEM;
 781
 782         return 0;
 783 }
 784
 785 void sd_zbc_release_disk(struct scsi_disk *sdkp)
 786 {
 787         kvfree(sdkp->zones_wp_offset);
 788         sdkp->zones_wp_offset = NULL;
 789         kfree(sdkp->zone_wp_update_buf);
 790         sdkp->zone_wp_update_buf = NULL;
 791 }