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 15 #include <asm/unaligned.h> 16 17 #include <scsi/scsi.h> 18 #include <scsi/scsi_cmnd.h> 19 20 #include "sd.h" 21 22 static int sd_zbc_parse_report(struct scsi_disk *sdkp, u8 *buf, 23 unsigned int idx, report_zones_cb cb, void *data) 24 { 25 struct scsi_device *sdp = sdkp->device; 26 struct blk_zone zone = { 0 }; 27 28 zone.type = buf[0] & 0x0f; 29 zone.cond = (buf[1] >> 4) & 0xf; 30 if (buf[1] & 0x01) 31 zone.reset = 1; 32 if (buf[1] & 0x02) 33 zone.non_seq = 1; 34 35 zone.len = logical_to_sectors(sdp, get_unaligned_be64(&buf[8])); 36 zone.start = logical_to_sectors(sdp, get_unaligned_be64(&buf[16])); 37 zone.wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24])); 38 if (zone.type != ZBC_ZONE_TYPE_CONV && 39 zone.cond == ZBC_ZONE_COND_FULL) 40 zone.wp = zone.start + zone.len; 41 42 return cb(&zone, idx, data); 43 } 44 45 /** 46 * sd_zbc_do_report_zones - Issue a REPORT ZONES scsi command. 47 * @sdkp: The target disk 48 * @buf: vmalloc-ed buffer to use for the reply 49 * @buflen: the buffer size 50 * @lba: Start LBA of the report 51 * @partial: Do partial report 52 * 53 * For internal use during device validation. 54 * Using partial=true can significantly speed up execution of a report zones 55 * command because the disk does not have to count all possible report matching 56 * zones and will only report the count of zones fitting in the command reply 57 * buffer. 58 */ 59 static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf, 60 unsigned int buflen, sector_t lba, 61 bool partial) 62 { 63 struct scsi_device *sdp = sdkp->device; 64 const int timeout = sdp->request_queue->rq_timeout; 65 struct scsi_sense_hdr sshdr; 66 unsigned char cmd[16]; 67 unsigned int rep_len; 68 int result; 69 70 memset(cmd, 0, 16); 71 cmd[0] = ZBC_IN; 72 cmd[1] = ZI_REPORT_ZONES; 73 put_unaligned_be64(lba, &cmd[2]); 74 put_unaligned_be32(buflen, &cmd[10]); 75 if (partial) 76 cmd[14] = ZBC_REPORT_ZONE_PARTIAL; 77 78 result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE, 79 buf, buflen, &sshdr, 80 timeout, SD_MAX_RETRIES, NULL); 81 if (result) { 82 sd_printk(KERN_ERR, sdkp, 83 "REPORT ZONES start lba %llu failed\n", lba); 84 sd_print_result(sdkp, "REPORT ZONES", result); 85 if (driver_byte(result) == DRIVER_SENSE && 86 scsi_sense_valid(&sshdr)) 87 sd_print_sense_hdr(sdkp, &sshdr); 88 return -EIO; 89 } 90 91 rep_len = get_unaligned_be32(&buf[0]); 92 if (rep_len < 64) { 93 sd_printk(KERN_ERR, sdkp, 94 "REPORT ZONES report invalid length %u\n", 95 rep_len); 96 return -EIO; 97 } 98 99 return 0; 100 } 101 102 /** 103 * Allocate a buffer for report zones reply. 104 * @sdkp: The target disk 105 * @nr_zones: Maximum number of zones to report 106 * @buflen: Size of the buffer allocated 107 * 108 * Try to allocate a reply buffer for the number of requested zones. 109 * The size of the buffer allocated may be smaller than requested to 110 * satify the device constraint (max_hw_sectors, max_segments, etc). 111 * 112 * Return the address of the allocated buffer and update @buflen with 113 * the size of the allocated buffer. 114 */ 115 static void *sd_zbc_alloc_report_buffer(struct scsi_disk *sdkp, 116 unsigned int nr_zones, size_t *buflen) 117 { 118 struct request_queue *q = sdkp->disk->queue; 119 size_t bufsize; 120 void *buf; 121 122 /* 123 * Report zone buffer size should be at most 64B times the number of 124 * zones requested plus the 64B reply header, but should be at least 125 * SECTOR_SIZE for ATA devices. 126 * Make sure that this size does not exceed the hardware capabilities. 127 * Furthermore, since the report zone command cannot be split, make 128 * sure that the allocated buffer can always be mapped by limiting the 129 * number of pages allocated to the HBA max segments limit. 130 */ 131 nr_zones = min(nr_zones, sdkp->nr_zones); 132 bufsize = roundup((nr_zones + 1) * 64, SECTOR_SIZE); 133 bufsize = min_t(size_t, bufsize, 134 queue_max_hw_sectors(q) << SECTOR_SHIFT); 135 bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT); 136 137 while (bufsize >= SECTOR_SIZE) { 138 buf = __vmalloc(bufsize, 139 GFP_KERNEL | __GFP_ZERO | __GFP_NORETRY); 140 if (buf) { 141 *buflen = bufsize; 142 return buf; 143 } 144 bufsize >>= 1; 145 } 146 147 return NULL; 148 } 149 150 /** 151 * sd_zbc_zone_sectors - Get the device zone size in number of 512B sectors. 152 * @sdkp: The target disk 153 */ 154 static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp) 155 { 156 return logical_to_sectors(sdkp->device, sdkp->zone_blocks); 157 } 158 159 int sd_zbc_report_zones(struct gendisk *disk, sector_t sector, 160 unsigned int nr_zones, report_zones_cb cb, void *data) 161 { 162 struct scsi_disk *sdkp = scsi_disk(disk); 163 sector_t capacity = logical_to_sectors(sdkp->device, sdkp->capacity); 164 unsigned int nr, i; 165 unsigned char *buf; 166 size_t offset, buflen = 0; 167 int zone_idx = 0; 168 int ret; 169 170 if (!sd_is_zoned(sdkp)) 171 /* Not a zoned device */ 172 return -EOPNOTSUPP; 173 174 if (!capacity) 175 /* Device gone or invalid */ 176 return -ENODEV; 177 178 buf = sd_zbc_alloc_report_buffer(sdkp, nr_zones, &buflen); 179 if (!buf) 180 return -ENOMEM; 181 182 while (zone_idx < nr_zones && sector < capacity) { 183 ret = sd_zbc_do_report_zones(sdkp, buf, buflen, 184 sectors_to_logical(sdkp->device, sector), true); 185 if (ret) 186 goto out; 187 188 offset = 0; 189 nr = min(nr_zones, get_unaligned_be32(&buf[0]) / 64); 190 if (!nr) 191 break; 192 193 for (i = 0; i < nr && zone_idx < nr_zones; i++) { 194 offset += 64; 195 ret = sd_zbc_parse_report(sdkp, buf + offset, zone_idx, 196 cb, data); 197 if (ret) 198 goto out; 199 zone_idx++; 200 } 201 202 sector += sd_zbc_zone_sectors(sdkp) * i; 203 } 204 205 ret = zone_idx; 206 out: 207 kvfree(buf); 208 return ret; 209 } 210 211 /** 212 * sd_zbc_setup_zone_mgmt_cmnd - Prepare a zone ZBC_OUT command. The operations 213 * can be RESET WRITE POINTER, OPEN, CLOSE or FINISH. 214 * @cmd: the command to setup 215 * @op: Operation to be performed 216 * @all: All zones control 217 * 218 * Called from sd_init_command() for REQ_OP_ZONE_RESET, REQ_OP_ZONE_RESET_ALL, 219 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE or REQ_OP_ZONE_FINISH requests. 220 */ 221 blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd, 222 unsigned char op, bool all) 223 { 224 struct request *rq = cmd->request; 225 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk); 226 sector_t sector = blk_rq_pos(rq); 227 sector_t block = sectors_to_logical(sdkp->device, sector); 228 229 if (!sd_is_zoned(sdkp)) 230 /* Not a zoned device */ 231 return BLK_STS_IOERR; 232 233 if (sdkp->device->changed) 234 return BLK_STS_IOERR; 235 236 if (sector & (sd_zbc_zone_sectors(sdkp) - 1)) 237 /* Unaligned request */ 238 return BLK_STS_IOERR; 239 240 cmd->cmd_len = 16; 241 memset(cmd->cmnd, 0, cmd->cmd_len); 242 cmd->cmnd[0] = ZBC_OUT; 243 cmd->cmnd[1] = op; 244 if (all) 245 cmd->cmnd[14] = 0x1; 246 else 247 put_unaligned_be64(block, &cmd->cmnd[2]); 248 249 rq->timeout = SD_TIMEOUT; 250 cmd->sc_data_direction = DMA_NONE; 251 cmd->transfersize = 0; 252 cmd->allowed = 0; 253 254 return BLK_STS_OK; 255 } 256 257 /** 258 * sd_zbc_complete - ZBC command post processing. 259 * @cmd: Completed command 260 * @good_bytes: Command reply bytes 261 * @sshdr: command sense header 262 * 263 * Called from sd_done(). Process report zones reply and handle reset zone 264 * and write commands errors. 265 */ 266 void sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes, 267 struct scsi_sense_hdr *sshdr) 268 { 269 int result = cmd->result; 270 struct request *rq = cmd->request; 271 272 if (op_is_zone_mgmt(req_op(rq)) && 273 result && 274 sshdr->sense_key == ILLEGAL_REQUEST && 275 sshdr->asc == 0x24) { 276 /* 277 * INVALID FIELD IN CDB error: a zone management command was 278 * attempted on a conventional zone. Nothing to worry about, 279 * so be quiet about the error. 280 */ 281 rq->rq_flags |= RQF_QUIET; 282 } 283 } 284 285 /** 286 * sd_zbc_check_zoned_characteristics - Check zoned block device characteristics 287 * @sdkp: Target disk 288 * @buf: Buffer where to store the VPD page data 289 * 290 * Read VPD page B6, get information and check that reads are unconstrained. 291 */ 292 static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp, 293 unsigned char *buf) 294 { 295 296 if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) { 297 sd_printk(KERN_NOTICE, sdkp, 298 "Read zoned characteristics VPD page failed\n"); 299 return -ENODEV; 300 } 301 302 if (sdkp->device->type != TYPE_ZBC) { 303 /* Host-aware */ 304 sdkp->urswrz = 1; 305 sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]); 306 sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]); 307 sdkp->zones_max_open = 0; 308 } else { 309 /* Host-managed */ 310 sdkp->urswrz = buf[4] & 1; 311 sdkp->zones_optimal_open = 0; 312 sdkp->zones_optimal_nonseq = 0; 313 sdkp->zones_max_open = get_unaligned_be32(&buf[16]); 314 } 315 316 /* 317 * Check for unconstrained reads: host-managed devices with 318 * constrained reads (drives failing read after write pointer) 319 * are not supported. 320 */ 321 if (!sdkp->urswrz) { 322 if (sdkp->first_scan) 323 sd_printk(KERN_NOTICE, sdkp, 324 "constrained reads devices are not supported\n"); 325 return -ENODEV; 326 } 327 328 return 0; 329 } 330 331 /** 332 * sd_zbc_check_capacity - Check the device capacity 333 * @sdkp: Target disk 334 * @buf: command buffer 335 * @zblock: zone size in number of blocks 336 * 337 * Get the device zone size and check that the device capacity as reported 338 * by READ CAPACITY matches the max_lba value (plus one) of the report zones 339 * command reply for devices with RC_BASIS == 0. 340 * 341 * Returns 0 upon success or an error code upon failure. 342 */ 343 static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf, 344 u32 *zblocks) 345 { 346 u64 zone_blocks; 347 sector_t max_lba; 348 unsigned char *rec; 349 int ret; 350 351 /* Do a report zone to get max_lba and the size of the first zone */ 352 ret = sd_zbc_do_report_zones(sdkp, buf, SD_BUF_SIZE, 0, false); 353 if (ret) 354 return ret; 355 356 if (sdkp->rc_basis == 0) { 357 /* The max_lba field is the capacity of this device */ 358 max_lba = get_unaligned_be64(&buf[8]); 359 if (sdkp->capacity != max_lba + 1) { 360 if (sdkp->first_scan) 361 sd_printk(KERN_WARNING, sdkp, 362 "Changing capacity from %llu to max LBA+1 %llu\n", 363 (unsigned long long)sdkp->capacity, 364 (unsigned long long)max_lba + 1); 365 sdkp->capacity = max_lba + 1; 366 } 367 } 368 369 /* Get the size of the first reported zone */ 370 rec = buf + 64; 371 zone_blocks = get_unaligned_be64(&rec[8]); 372 if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) { 373 if (sdkp->first_scan) 374 sd_printk(KERN_NOTICE, sdkp, 375 "Zone size too large\n"); 376 return -EFBIG; 377 } 378 379 *zblocks = zone_blocks; 380 381 return 0; 382 } 383 384 int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf) 385 { 386 struct gendisk *disk = sdkp->disk; 387 unsigned int nr_zones; 388 u32 zone_blocks = 0; 389 int ret; 390 391 if (!sd_is_zoned(sdkp)) 392 /* 393 * Device managed or normal SCSI disk, 394 * no special handling required 395 */ 396 return 0; 397 398 /* Check zoned block device characteristics (unconstrained reads) */ 399 ret = sd_zbc_check_zoned_characteristics(sdkp, buf); 400 if (ret) 401 goto err; 402 403 /* Check the device capacity reported by report zones */ 404 ret = sd_zbc_check_capacity(sdkp, buf, &zone_blocks); 405 if (ret != 0) 406 goto err; 407 408 /* The drive satisfies the kernel restrictions: set it up */ 409 blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, sdkp->disk->queue); 410 blk_queue_required_elevator_features(sdkp->disk->queue, 411 ELEVATOR_F_ZBD_SEQ_WRITE); 412 nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks); 413 414 /* READ16/WRITE16 is mandatory for ZBC disks */ 415 sdkp->device->use_16_for_rw = 1; 416 sdkp->device->use_10_for_rw = 0; 417 418 /* 419 * Revalidate the disk zone bitmaps once the block device capacity is 420 * set on the second revalidate execution during disk scan and if 421 * something changed when executing a normal revalidate. 422 */ 423 if (sdkp->first_scan) { 424 sdkp->zone_blocks = zone_blocks; 425 sdkp->nr_zones = nr_zones; 426 return 0; 427 } 428 429 if (sdkp->zone_blocks != zone_blocks || 430 sdkp->nr_zones != nr_zones || 431 disk->queue->nr_zones != nr_zones) { 432 ret = blk_revalidate_disk_zones(disk); 433 if (ret != 0) 434 goto err; 435 sdkp->zone_blocks = zone_blocks; 436 sdkp->nr_zones = nr_zones; 437 } 438 439 return 0; 440 441 err: 442 sdkp->capacity = 0; 443 444 return ret; 445 } 446 447 void sd_zbc_print_zones(struct scsi_disk *sdkp) 448 { 449 if (!sd_is_zoned(sdkp) || !sdkp->capacity) 450 return; 451 452 if (sdkp->capacity & (sdkp->zone_blocks - 1)) 453 sd_printk(KERN_NOTICE, sdkp, 454 "%u zones of %u logical blocks + 1 runt zone\n", 455 sdkp->nr_zones - 1, 456 sdkp->zone_blocks); 457 else 458 sd_printk(KERN_NOTICE, sdkp, 459 "%u zones of %u logical blocks\n", 460 sdkp->nr_zones, 461 sdkp->zone_blocks); 462 } 463