drivers/scsi/sd_zbc.c

   1 /*
   2  * SCSI Zoned Block commands
   3  *
   4  * Copyright (C) 2014-2015 SUSE Linux GmbH
   5  * Written by: Hannes Reinecke <hare@suse.de>
   6  * Modified by: Damien Le Moal <damien.lemoal@hgst.com>
   7  * Modified by: Shaun Tancheff <shaun.tancheff@seagate.com>
   8  *
   9  * This program is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU General Public License version
  11  * 2 as published by the Free Software Foundation.
  12  *
  13  * This program is distributed in the hope that it will be useful, but
  14  * WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * along with this program; see the file COPYING.  If not, write to
  20  * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
  21  * USA.
  22  *
  23  */
  24
  25 #include <linux/blkdev.h>
  26
  27 #include <asm/unaligned.h>
  28
  29 #include <scsi/scsi.h>
  30 #include <scsi/scsi_cmnd.h>
  31 #include <scsi/scsi_dbg.h>
  32 #include <scsi/scsi_device.h>
  33 #include <scsi/scsi_driver.h>
  34 #include <scsi/scsi_host.h>
  35 #include <scsi/scsi_eh.h>
  36
  37 #include "sd.h"
  38 #include "scsi_priv.h"
  39
  40 enum zbc_zone_type {
  41         ZBC_ZONE_TYPE_CONV = 0x1,
  42         ZBC_ZONE_TYPE_SEQWRITE_REQ,
  43         ZBC_ZONE_TYPE_SEQWRITE_PREF,
  44         ZBC_ZONE_TYPE_RESERVED,
  45 };
  46
  47 enum zbc_zone_cond {
  48         ZBC_ZONE_COND_NO_WP,
  49         ZBC_ZONE_COND_EMPTY,
  50         ZBC_ZONE_COND_IMP_OPEN,
  51         ZBC_ZONE_COND_EXP_OPEN,
  52         ZBC_ZONE_COND_CLOSED,
  53         ZBC_ZONE_COND_READONLY = 0xd,
  54         ZBC_ZONE_COND_FULL,
  55         ZBC_ZONE_COND_OFFLINE,
  56 };
  57
  58 /**
  59  * Convert a zone descriptor to a zone struct.
  60  */
  61 static void sd_zbc_parse_report(struct scsi_disk *sdkp,
  62                                 u8 *buf,
  63                                 struct blk_zone *zone)
  64 {
  65         struct scsi_device *sdp = sdkp->device;
  66
  67         memset(zone, 0, sizeof(struct blk_zone));
  68
  69         zone->type = buf[0] & 0x0f;
  70         zone->cond = (buf[1] >> 4) & 0xf;
  71         if (buf[1] & 0x01)
  72                 zone->reset = 1;
  73         if (buf[1] & 0x02)
  74                 zone->non_seq = 1;
  75
  76         zone->len = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
  77         zone->start = logical_to_sectors(sdp, get_unaligned_be64(&buf[16]));
  78         zone->wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24]));
  79         if (zone->type != ZBC_ZONE_TYPE_CONV &&
  80             zone->cond == ZBC_ZONE_COND_FULL)
  81                 zone->wp = zone->start + zone->len;
  82 }
  83
  84 /**
  85  * Issue a REPORT ZONES scsi command.
  86  */
  87 static int sd_zbc_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
  88                                unsigned int buflen, sector_t lba)
  89 {
  90         struct scsi_device *sdp = sdkp->device;
  91         const int timeout = sdp->request_queue->rq_timeout;
  92         struct scsi_sense_hdr sshdr;
  93         unsigned char cmd[16];
  94         unsigned int rep_len;
  95         int result;
  96
  97         memset(cmd, 0, 16);
  98         cmd[0] = ZBC_IN;
  99         cmd[1] = ZI_REPORT_ZONES;
 100         put_unaligned_be64(lba, &cmd[2]);
 101         put_unaligned_be32(buflen, &cmd[10]);
 102         memset(buf, 0, buflen);
 103
 104         result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
 105                                   buf, buflen, &sshdr,
 106                                   timeout, SD_MAX_RETRIES, NULL);
 107         if (result) {
 108                 sd_printk(KERN_ERR, sdkp,
 109                           "REPORT ZONES lba %llu failed with %d/%d\n",
 110                           (unsigned long long)lba,
 111                           host_byte(result), driver_byte(result));
 112                 return -EIO;
 113         }
 114
 115         rep_len = get_unaligned_be32(&buf[0]);
 116         if (rep_len < 64) {
 117                 sd_printk(KERN_ERR, sdkp,
 118                           "REPORT ZONES report invalid length %u\n",
 119                           rep_len);
 120                 return -EIO;
 121         }
 122
 123         return 0;
 124 }
 125
 126 int sd_zbc_setup_report_cmnd(struct scsi_cmnd *cmd)
 127 {
 128         struct request *rq = cmd->request;
 129         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 130         sector_t lba, sector = blk_rq_pos(rq);
 131         unsigned int nr_bytes = blk_rq_bytes(rq);
 132         int ret;
 133
 134         WARN_ON(nr_bytes == 0);
 135
 136         if (!sd_is_zoned(sdkp))
 137                 /* Not a zoned device */
 138                 return BLKPREP_KILL;
 139
 140         ret = scsi_init_io(cmd);
 141         if (ret != BLKPREP_OK)
 142                 return ret;
 143
 144         cmd->cmd_len = 16;
 145         memset(cmd->cmnd, 0, cmd->cmd_len);
 146         cmd->cmnd[0] = ZBC_IN;
 147         cmd->cmnd[1] = ZI_REPORT_ZONES;
 148         lba = sectors_to_logical(sdkp->device, sector);
 149         put_unaligned_be64(lba, &cmd->cmnd[2]);
 150         put_unaligned_be32(nr_bytes, &cmd->cmnd[10]);
 151         /* Do partial report for speeding things up */
 152         cmd->cmnd[14] = ZBC_REPORT_ZONE_PARTIAL;
 153
 154         cmd->sc_data_direction = DMA_FROM_DEVICE;
 155         cmd->sdb.length = nr_bytes;
 156         cmd->transfersize = sdkp->device->sector_size;
 157         cmd->allowed = 0;
 158
 159         /*
 160          * Report may return less bytes than requested. Make sure
 161          * to report completion on the entire initial request.
 162          */
 163         rq->__data_len = nr_bytes;
 164
 165         return BLKPREP_OK;
 166 }
 167
 168 static void sd_zbc_report_zones_complete(struct scsi_cmnd *scmd,
 169                                          unsigned int good_bytes)
 170 {
 171         struct request *rq = scmd->request;
 172         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 173         struct sg_mapping_iter miter;
 174         struct blk_zone_report_hdr hdr;
 175         struct blk_zone zone;
 176         unsigned int offset, bytes = 0;
 177         unsigned long flags;
 178         u8 *buf;
 179
 180         if (good_bytes < 64)
 181                 return;
 182
 183         memset(&hdr, 0, sizeof(struct blk_zone_report_hdr));
 184
 185         sg_miter_start(&miter, scsi_sglist(scmd), scsi_sg_count(scmd),
 186                        SG_MITER_TO_SG | SG_MITER_ATOMIC);
 187
 188         local_irq_save(flags);
 189         while (sg_miter_next(&miter) && bytes < good_bytes) {
 190
 191                 buf = miter.addr;
 192                 offset = 0;
 193
 194                 if (bytes == 0) {
 195                         /* Set the report header */
 196                         hdr.nr_zones = min_t(unsigned int,
 197                                          (good_bytes - 64) / 64,
 198                                          get_unaligned_be32(&buf[0]) / 64);
 199                         memcpy(buf, &hdr, sizeof(struct blk_zone_report_hdr));
 200                         offset += 64;
 201                         bytes += 64;
 202                 }
 203
 204                 /* Parse zone descriptors */
 205                 while (offset < miter.length && hdr.nr_zones) {
 206                         WARN_ON(offset > miter.length);
 207                         buf = miter.addr + offset;
 208                         sd_zbc_parse_report(sdkp, buf, &zone);
 209                         memcpy(buf, &zone, sizeof(struct blk_zone));
 210                         offset += 64;
 211                         bytes += 64;
 212                         hdr.nr_zones--;
 213                 }
 214
 215                 if (!hdr.nr_zones)
 216                         break;
 217
 218         }
 219         sg_miter_stop(&miter);
 220         local_irq_restore(flags);
 221 }
 222
 223 static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
 224 {
 225         return logical_to_sectors(sdkp->device, sdkp->zone_blocks);
 226 }
 227
 228 static inline unsigned int sd_zbc_zone_no(struct scsi_disk *sdkp,
 229                                           sector_t sector)
 230 {
 231         return sectors_to_logical(sdkp->device, sector) >> sdkp->zone_shift;
 232 }
 233
 234 int sd_zbc_setup_reset_cmnd(struct scsi_cmnd *cmd)
 235 {
 236         struct request *rq = cmd->request;
 237         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 238         sector_t sector = blk_rq_pos(rq);
 239         sector_t block = sectors_to_logical(sdkp->device, sector);
 240         unsigned int zno = block >> sdkp->zone_shift;
 241
 242         if (!sd_is_zoned(sdkp))
 243                 /* Not a zoned device */
 244                 return BLKPREP_KILL;
 245
 246         if (sdkp->device->changed)
 247                 return BLKPREP_KILL;
 248
 249         if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
 250                 /* Unaligned request */
 251                 return BLKPREP_KILL;
 252
 253         /* Do not allow concurrent reset and writes */
 254         if (sdkp->zones_wlock &&
 255             test_and_set_bit(zno, sdkp->zones_wlock))
 256                 return BLKPREP_DEFER;
 257
 258         cmd->cmd_len = 16;
 259         memset(cmd->cmnd, 0, cmd->cmd_len);
 260         cmd->cmnd[0] = ZBC_OUT;
 261         cmd->cmnd[1] = ZO_RESET_WRITE_POINTER;
 262         put_unaligned_be64(block, &cmd->cmnd[2]);
 263
 264         rq->timeout = SD_TIMEOUT;
 265         cmd->sc_data_direction = DMA_NONE;
 266         cmd->transfersize = 0;
 267         cmd->allowed = 0;
 268
 269         return BLKPREP_OK;
 270 }
 271
 272 int sd_zbc_write_lock_zone(struct scsi_cmnd *cmd)
 273 {
 274         struct request *rq = cmd->request;
 275         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 276         sector_t sector = blk_rq_pos(rq);
 277         sector_t zone_sectors = sd_zbc_zone_sectors(sdkp);
 278         unsigned int zno = sd_zbc_zone_no(sdkp, sector);
 279
 280         /*
 281          * Note: Checks of the alignment of the write command on
 282          * logical blocks is done in sd.c
 283          */
 284
 285         /* Do not allow zone boundaries crossing on host-managed drives */
 286         if (blk_queue_zoned_model(sdkp->disk->queue) == BLK_ZONED_HM &&
 287             (sector & (zone_sectors - 1)) + blk_rq_sectors(rq) > zone_sectors)
 288                 return BLKPREP_KILL;
 289
 290         /*
 291          * Do not issue more than one write at a time per
 292          * zone. This solves write ordering problems due to
 293          * the unlocking of the request queue in the dispatch
 294          * path in the non scsi-mq case. For scsi-mq, this
 295          * also avoids potential write reordering when multiple
 296          * threads running on different CPUs write to the same
 297          * zone (with a synchronized sequential pattern).
 298          */
 299         if (sdkp->zones_wlock &&
 300             test_and_set_bit(zno, sdkp->zones_wlock))
 301                 return BLKPREP_DEFER;
 302
 303         return BLKPREP_OK;
 304 }
 305
 306 void sd_zbc_write_unlock_zone(struct scsi_cmnd *cmd)
 307 {
 308         struct request *rq = cmd->request;
 309         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 310
 311         if (sdkp->zones_wlock) {
 312                 unsigned int zno = sd_zbc_zone_no(sdkp, blk_rq_pos(rq));
 313                 WARN_ON_ONCE(!test_bit(zno, sdkp->zones_wlock));
 314                 clear_bit_unlock(zno, sdkp->zones_wlock);
 315                 smp_mb__after_atomic();
 316         }
 317 }
 318
 319 void sd_zbc_complete(struct scsi_cmnd *cmd,
 320                      unsigned int good_bytes,
 321                      struct scsi_sense_hdr *sshdr)
 322 {
 323         int result = cmd->result;
 324         struct request *rq = cmd->request;
 325
 326         switch (req_op(rq)) {
 327         case REQ_OP_WRITE:
 328         case REQ_OP_WRITE_SAME:
 329         case REQ_OP_ZONE_RESET:
 330
 331                 /* Unlock the zone */
 332                 sd_zbc_write_unlock_zone(cmd);
 333
 334                 if (!result ||
 335                     sshdr->sense_key != ILLEGAL_REQUEST)
 336                         break;
 337
 338                 switch (sshdr->asc) {
 339                 case 0x24:
 340                         /*
 341                          * INVALID FIELD IN CDB error: For a zone reset,
 342                          * this means that a reset of a conventional
 343                          * zone was attempted. Nothing to worry about in
 344                          * this case, so be quiet about the error.
 345                          */
 346                         if (req_op(rq) == REQ_OP_ZONE_RESET)
 347                                 rq->rq_flags |= RQF_QUIET;
 348                         break;
 349                 case 0x21:
 350                         /*
 351                          * INVALID ADDRESS FOR WRITE error: It is unlikely that
 352                          * retrying write requests failed with any kind of
 353                          * alignement error will result in success. So don't.
 354                          */
 355                         cmd->allowed = 0;
 356                         break;
 357                 }
 358
 359                 break;
 360
 361         case REQ_OP_ZONE_REPORT:
 362
 363                 if (!result)
 364                         sd_zbc_report_zones_complete(cmd, good_bytes);
 365                 break;
 366
 367         }
 368 }
 369
 370 /**
 371  * Read zoned block device characteristics (VPD page B6).
 372  */
 373 static int sd_zbc_read_zoned_characteristics(struct scsi_disk *sdkp,
 374                                              unsigned char *buf)
 375 {
 376
 377         if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
 378                 sd_printk(KERN_NOTICE, sdkp,
 379                           "Unconstrained-read check failed\n");
 380                 return -ENODEV;
 381         }
 382
 383         if (sdkp->device->type != TYPE_ZBC) {
 384                 /* Host-aware */
 385                 sdkp->urswrz = 1;
 386                 sdkp->zones_optimal_open = get_unaligned_be64(&buf[8]);
 387                 sdkp->zones_optimal_nonseq = get_unaligned_be64(&buf[12]);
 388                 sdkp->zones_max_open = 0;
 389         } else {
 390                 /* Host-managed */
 391                 sdkp->urswrz = buf[4] & 1;
 392                 sdkp->zones_optimal_open = 0;
 393                 sdkp->zones_optimal_nonseq = 0;
 394                 sdkp->zones_max_open = get_unaligned_be64(&buf[16]);
 395         }
 396
 397         return 0;
 398 }
 399
 400 /**
 401  * Check reported capacity.
 402  */
 403 static int sd_zbc_check_capacity(struct scsi_disk *sdkp,
 404                                  unsigned char *buf)
 405 {
 406         sector_t lba;
 407         int ret;
 408
 409         if (sdkp->rc_basis != 0)
 410                 return 0;
 411
 412         /* Do a report zone to get the maximum LBA to check capacity */
 413         ret = sd_zbc_report_zones(sdkp, buf, SD_BUF_SIZE, 0);
 414         if (ret)
 415                 return ret;
 416
 417         /* The max_lba field is the capacity of this device */
 418         lba = get_unaligned_be64(&buf[8]);
 419         if (lba + 1 == sdkp->capacity)
 420                 return 0;
 421
 422         if (sdkp->first_scan)
 423                 sd_printk(KERN_WARNING, sdkp,
 424                           "Changing capacity from %llu to max LBA+1 %llu\n",
 425                           (unsigned long long)sdkp->capacity,
 426                           (unsigned long long)lba + 1);
 427         sdkp->capacity = lba + 1;
 428
 429         return 0;
 430 }
 431
 432 #define SD_ZBC_BUF_SIZE 131072
 433
 434 static int sd_zbc_check_zone_size(struct scsi_disk *sdkp)
 435 {
 436         u64 zone_blocks;
 437         sector_t block = 0;
 438         unsigned char *buf;
 439         unsigned char *rec;
 440         unsigned int buf_len;
 441         unsigned int list_length;
 442         int ret;
 443         u8 same;
 444
 445         sdkp->zone_blocks = 0;
 446
 447         /* Get a buffer */
 448         buf = kmalloc(SD_ZBC_BUF_SIZE, GFP_KERNEL);
 449         if (!buf)
 450                 return -ENOMEM;
 451
 452         /* Do a report zone to get the same field */
 453         ret = sd_zbc_report_zones(sdkp, buf, SD_ZBC_BUF_SIZE, 0);
 454         if (ret) {
 455                 zone_blocks = 0;
 456                 goto out;
 457         }
 458
 459         same = buf[4] & 0x0f;
 460         if (same > 0) {
 461                 rec = &buf[64];
 462                 zone_blocks = get_unaligned_be64(&rec[8]);
 463                 goto out;
 464         }
 465
 466         /*
 467          * Check the size of all zones: all zones must be of
 468          * equal size, except the last zone which can be smaller
 469          * than other zones.
 470          */
 471         do {
 472
 473                 /* Parse REPORT ZONES header */
 474                 list_length = get_unaligned_be32(&buf[0]) + 64;
 475                 rec = buf + 64;
 476                 if (list_length < SD_ZBC_BUF_SIZE)
 477                         buf_len = list_length;
 478                 else
 479                         buf_len = SD_ZBC_BUF_SIZE;
 480
 481                 /* Parse zone descriptors */
 482                 while (rec < buf + buf_len) {
 483                         zone_blocks = get_unaligned_be64(&rec[8]);
 484                         if (sdkp->zone_blocks == 0) {
 485                                 sdkp->zone_blocks = zone_blocks;
 486                         } else if (zone_blocks != sdkp->zone_blocks &&
 487                                    (block + zone_blocks < sdkp->capacity
 488                                     || zone_blocks > sdkp->zone_blocks)) {
 489                                 zone_blocks = 0;
 490                                 goto out;
 491                         }
 492                         block += zone_blocks;
 493                         rec += 64;
 494                 }
 495
 496                 if (block < sdkp->capacity) {
 497                         ret = sd_zbc_report_zones(sdkp, buf,
 498                                                   SD_ZBC_BUF_SIZE, block);
 499                         if (ret)
 500                                 return ret;
 501                 }
 502
 503         } while (block < sdkp->capacity);
 504
 505         zone_blocks = sdkp->zone_blocks;
 506
 507 out:
 508         kfree(buf);
 509
 510         if (!zone_blocks) {
 511                 if (sdkp->first_scan)
 512                         sd_printk(KERN_NOTICE, sdkp,
 513                                   "Devices with non constant zone "
 514                                   "size are not supported\n");
 515                 return -ENODEV;
 516         }
 517
 518         if (!is_power_of_2(zone_blocks)) {
 519                 if (sdkp->first_scan)
 520                         sd_printk(KERN_NOTICE, sdkp,
 521                                   "Devices with non power of 2 zone "
 522                                   "size are not supported\n");
 523                 return -ENODEV;
 524         }
 525
 526         if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
 527                 if (sdkp->first_scan)
 528                         sd_printk(KERN_NOTICE, sdkp,
 529                                   "Zone size too large\n");
 530                 return -ENODEV;
 531         }
 532
 533         sdkp->zone_blocks = zone_blocks;
 534
 535         return 0;
 536 }
 537
 538 static int sd_zbc_setup(struct scsi_disk *sdkp)
 539 {
 540
 541         /* chunk_sectors indicates the zone size */
 542         blk_queue_chunk_sectors(sdkp->disk->queue,
 543                         logical_to_sectors(sdkp->device, sdkp->zone_blocks));
 544         sdkp->zone_shift = ilog2(sdkp->zone_blocks);
 545         sdkp->nr_zones = sdkp->capacity >> sdkp->zone_shift;
 546         if (sdkp->capacity & (sdkp->zone_blocks - 1))
 547                 sdkp->nr_zones++;
 548
 549         if (!sdkp->zones_wlock) {
 550                 sdkp->zones_wlock = kcalloc(BITS_TO_LONGS(sdkp->nr_zones),
 551                                             sizeof(unsigned long),
 552                                             GFP_KERNEL);
 553                 if (!sdkp->zones_wlock)
 554                         return -ENOMEM;
 555         }
 556
 557         return 0;
 558 }
 559
 560 int sd_zbc_read_zones(struct scsi_disk *sdkp,
 561                       unsigned char *buf)
 562 {
 563         int ret;
 564
 565         if (!sd_is_zoned(sdkp))
 566                 /*
 567                  * Device managed or normal SCSI disk,
 568                  * no special handling required
 569                  */
 570                 return 0;
 571
 572
 573         /* Get zoned block device characteristics */
 574         ret = sd_zbc_read_zoned_characteristics(sdkp, buf);
 575         if (ret)
 576                 goto err;
 577
 578         /*
 579          * Check for unconstrained reads: host-managed devices with
 580          * constrained reads (drives failing read after write pointer)
 581          * are not supported.
 582          */
 583         if (!sdkp->urswrz) {
 584                 if (sdkp->first_scan)
 585                         sd_printk(KERN_NOTICE, sdkp,
 586                           "constrained reads devices are not supported\n");
 587                 ret = -ENODEV;
 588                 goto err;
 589         }
 590
 591         /* Check capacity */
 592         ret = sd_zbc_check_capacity(sdkp, buf);
 593         if (ret)
 594                 goto err;
 595
 596         /*
 597          * Check zone size: only devices with a constant zone size (except
 598          * an eventual last runt zone) that is a power of 2 are supported.
 599          */
 600         ret = sd_zbc_check_zone_size(sdkp);
 601         if (ret)
 602                 goto err;
 603
 604         /* The drive satisfies the kernel restrictions: set it up */
 605         ret = sd_zbc_setup(sdkp);
 606         if (ret)
 607                 goto err;
 608
 609         /* READ16/WRITE16 is mandatory for ZBC disks */
 610         sdkp->device->use_16_for_rw = 1;
 611         sdkp->device->use_10_for_rw = 0;
 612
 613         return 0;
 614
 615 err:
 616         sdkp->capacity = 0;
 617
 618         return ret;
 619 }
 620
 621 void sd_zbc_remove(struct scsi_disk *sdkp)
 622 {
 623         kfree(sdkp->zones_wlock);
 624         sdkp->zones_wlock = NULL;
 625 }
 626
 627 void sd_zbc_print_zones(struct scsi_disk *sdkp)
 628 {
 629         if (!sd_is_zoned(sdkp) || !sdkp->capacity)
 630                 return;
 631
 632         if (sdkp->capacity & (sdkp->zone_blocks - 1))
 633                 sd_printk(KERN_NOTICE, sdkp,
 634                           "%u zones of %u logical blocks + 1 runt zone\n",
 635                           sdkp->nr_zones - 1,
 636                           sdkp->zone_blocks);
 637         else
 638                 sd_printk(KERN_NOTICE, sdkp,
 639                           "%u zones of %u logical blocks\n",
 640                           sdkp->nr_zones,
 641                           sdkp->zone_blocks);
 642 }