1 // SPDX-License-Identifier: GPL-2.0-only
3 * SCSI Zoned Block commands
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>
11 #include <linux/blkdev.h>
12 #include <linux/vmalloc.h>
13 #include <linux/sched/mm.h>
14 #include <linux/mutex.h>
16 #include <asm/unaligned.h>
18 #include <scsi/scsi.h>
19 #include <scsi/scsi_cmnd.h>
23 static unsigned int sd_zbc_get_zone_wp_offset(struct blk_zone *zone)
25 if (zone->type == ZBC_ZONE_TYPE_CONV)
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:
35 case BLK_ZONE_COND_EMPTY:
36 case BLK_ZONE_COND_OFFLINE:
37 case BLK_ZONE_COND_READONLY:
40 * Offline and read-only zones do not have a valid
41 * write pointer. Use 0 as for an empty zone.
47 static int sd_zbc_parse_report(struct scsi_disk *sdkp, u8 *buf,
48 unsigned int idx, report_zones_cb cb, void *data)
50 struct scsi_device *sdp = sdkp->device;
51 struct blk_zone zone = { 0 };
54 zone.type = buf[0] & 0x0f;
55 zone.cond = (buf[1] >> 4) & 0xf;
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;
68 ret = cb(&zone, idx, data);
72 if (sdkp->rev_wp_offset)
73 sdkp->rev_wp_offset[idx] = sd_zbc_get_zone_wp_offset(&zone);
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
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
92 static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
93 unsigned int buflen, sector_t lba,
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;
105 cmd[1] = ZI_REPORT_ZONES;
106 put_unaligned_be64(lba, &cmd[2]);
107 put_unaligned_be32(buflen, &cmd[10]);
109 cmd[14] = ZBC_REPORT_ZONE_PARTIAL;
111 result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
113 timeout, SD_MAX_RETRIES, NULL);
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);
124 rep_len = get_unaligned_be32(&buf[0]);
126 sd_printk(KERN_ERR, sdkp,
127 "REPORT ZONES report invalid length %u\n",
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
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).
145 * Return the address of the allocated buffer and update @buflen with
146 * the size of the allocated buffer.
148 static void *sd_zbc_alloc_report_buffer(struct scsi_disk *sdkp,
149 unsigned int nr_zones, size_t *buflen)
151 struct request_queue *q = sdkp->disk->queue;
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.
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);
170 while (bufsize >= SECTOR_SIZE) {
171 buf = __vmalloc(bufsize,
172 GFP_KERNEL | __GFP_ZERO | __GFP_NORETRY);
184 * sd_zbc_zone_sectors - Get the device zone size in number of 512B sectors.
185 * @sdkp: The target disk
187 static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
189 return logical_to_sectors(sdkp->device, sdkp->zone_blocks);
192 int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
193 unsigned int nr_zones, report_zones_cb cb, void *data)
195 struct scsi_disk *sdkp = scsi_disk(disk);
196 sector_t capacity = logical_to_sectors(sdkp->device, sdkp->capacity);
199 size_t offset, buflen = 0;
203 if (!sd_is_zoned(sdkp))
204 /* Not a zoned device */
208 /* Device gone or invalid */
211 buf = sd_zbc_alloc_report_buffer(sdkp, nr_zones, &buflen);
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);
222 nr = min(nr_zones, get_unaligned_be32(&buf[0]) / 64);
226 for (i = 0; i < nr && zone_idx < nr_zones; i++) {
228 ret = sd_zbc_parse_report(sdkp, buf + offset, zone_idx,
235 sector += sd_zbc_zone_sectors(sdkp) * i;
244 static blk_status_t sd_zbc_cmnd_checks(struct scsi_cmnd *cmd)
246 struct request *rq = cmd->request;
247 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
248 sector_t sector = blk_rq_pos(rq);
250 if (!sd_is_zoned(sdkp))
251 /* Not a zoned device */
252 return BLK_STS_IOERR;
254 if (sdkp->device->changed)
255 return BLK_STS_IOERR;
257 if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
258 /* Unaligned request */
259 return BLK_STS_IOERR;
264 #define SD_ZBC_INVALID_WP_OFST (~0u)
265 #define SD_ZBC_UPDATING_WP_OFST (SD_ZBC_INVALID_WP_OFST - 1)
267 static int sd_zbc_update_wp_offset_cb(struct blk_zone *zone, unsigned int idx,
270 struct scsi_disk *sdkp = data;
272 lockdep_assert_held(&sdkp->zones_wp_offset_lock);
274 sdkp->zones_wp_offset[idx] = sd_zbc_get_zone_wp_offset(zone);
279 static void sd_zbc_update_wp_offset_workfn(struct work_struct *work)
281 struct scsi_disk *sdkp;
285 sdkp = container_of(work, struct scsi_disk, zone_wp_offset_work);
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)
292 spin_unlock_bh(&sdkp->zones_wp_offset_lock);
293 ret = sd_zbc_do_report_zones(sdkp, sdkp->zone_wp_update_buf,
295 zno * sdkp->zone_blocks, true);
296 spin_lock_bh(&sdkp->zones_wp_offset_lock);
298 sd_zbc_parse_report(sdkp, sdkp->zone_wp_update_buf + 64,
299 zno, sd_zbc_update_wp_offset_cb,
302 spin_unlock_bh(&sdkp->zones_wp_offset_lock);
304 scsi_device_put(sdkp->device);
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
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.
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.
320 blk_status_t sd_zbc_prepare_zone_append(struct scsi_cmnd *cmd, sector_t *lba,
321 unsigned int nr_blocks)
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);
328 ret = sd_zbc_cmnd_checks(cmd);
329 if (ret != BLK_STS_OK)
332 if (!blk_rq_zone_is_seq(rq))
333 return BLK_STS_IOERR;
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;
339 spin_lock_bh(&sdkp->zones_wp_offset_lock);
340 wp_offset = sdkp->zones_wp_offset[zno];
342 case SD_ZBC_INVALID_WP_OFST:
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.
349 if (scsi_device_get(sdkp->device)) {
353 sdkp->zones_wp_offset[zno] = SD_ZBC_UPDATING_WP_OFST;
354 schedule_work(&sdkp->zone_wp_offset_work);
356 case SD_ZBC_UPDATING_WP_OFST:
357 ret = BLK_STS_DEV_RESOURCE;
360 wp_offset = sectors_to_logical(sdkp->device, wp_offset);
361 if (wp_offset + nr_blocks > sdkp->zone_blocks) {
368 spin_unlock_bh(&sdkp->zones_wp_offset_lock);
370 blk_req_zone_write_unlock(rq);
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
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.
384 blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
385 unsigned char op, bool all)
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);
393 ret = sd_zbc_cmnd_checks(cmd);
394 if (ret != BLK_STS_OK)
398 memset(cmd->cmnd, 0, cmd->cmd_len);
399 cmd->cmnd[0] = ZBC_OUT;
404 put_unaligned_be64(block, &cmd->cmnd[2]);
406 rq->timeout = SD_TIMEOUT;
407 cmd->sc_data_direction = DMA_NONE;
408 cmd->transfersize = 0;
414 static bool sd_zbc_need_zone_wp_update(struct request *rq)
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:
423 case REQ_OP_WRITE_ZEROES:
424 case REQ_OP_WRITE_SAME:
425 return blk_rq_zone_is_seq(rq);
432 * sd_zbc_zone_wp_update - Update cached zone write pointer upon cmd completion
433 * @cmd: Completed command
434 * @good_bytes: Command reply bytes
436 * Called from sd_zbc_complete() to handle the update of the cached zone write
437 * pointer value in case an update is needed.
439 static unsigned int sd_zbc_zone_wp_update(struct scsi_cmnd *cmd,
440 unsigned int good_bytes)
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);
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
454 spin_lock_bh(&sdkp->zones_wp_offset_lock);
456 if (result && op != REQ_OP_ZONE_RESET_ALL) {
457 if (op == REQ_OP_ZONE_APPEND) {
458 /* Force complete completion (no retry) */
460 scsi_set_resid(cmd, blk_rq_bytes(rq));
464 * Force an update of the zone write pointer offset on
465 * the next zone append access.
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;
473 case REQ_OP_ZONE_APPEND:
474 rq->__sector += sdkp->zones_wp_offset[zno];
476 case REQ_OP_WRITE_ZEROES:
477 case REQ_OP_WRITE_SAME:
479 if (sdkp->zones_wp_offset[zno] < sd_zbc_zone_sectors(sdkp))
480 sdkp->zones_wp_offset[zno] +=
481 good_bytes >> SECTOR_SHIFT;
483 case REQ_OP_ZONE_RESET:
484 sdkp->zones_wp_offset[zno] = 0;
486 case REQ_OP_ZONE_FINISH:
487 sdkp->zones_wp_offset[zno] = sd_zbc_zone_sectors(sdkp);
489 case REQ_OP_ZONE_RESET_ALL:
490 memset(sdkp->zones_wp_offset, 0,
491 sdkp->nr_zones * sizeof(unsigned int));
498 spin_unlock_bh(&sdkp->zones_wp_offset_lock);
504 * sd_zbc_complete - ZBC command post processing.
505 * @cmd: Completed command
506 * @good_bytes: Command reply bytes
507 * @sshdr: command sense header
509 * Called from sd_done() to handle zone commands errors and updates to the
510 * device queue zone write pointer offset cahce.
512 unsigned int sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
513 struct scsi_sense_hdr *sshdr)
515 int result = cmd->result;
516 struct request *rq = cmd->request;
518 if (op_is_zone_mgmt(req_op(rq)) &&
520 sshdr->sense_key == ILLEGAL_REQUEST &&
521 sshdr->asc == 0x24) {
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.
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);
531 if (req_op(rq) == REQ_OP_ZONE_APPEND)
532 blk_req_zone_write_unlock(rq);
538 * sd_zbc_check_zoned_characteristics - Check zoned block device characteristics
540 * @buf: Buffer where to store the VPD page data
542 * Read VPD page B6, get information and check that reads are unconstrained.
544 static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
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");
554 if (sdkp->device->type != TYPE_ZBC) {
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;
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]);
569 * Check for unconstrained reads: host-managed devices with
570 * constrained reads (drives failing read after write pointer)
574 if (sdkp->first_scan)
575 sd_printk(KERN_NOTICE, sdkp,
576 "constrained reads devices are not supported\n");
584 * sd_zbc_check_capacity - Check the device capacity
586 * @buf: command buffer
587 * @zblocks: zone size in number of blocks
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.
593 * Returns 0 upon success or an error code upon failure.
595 static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf,
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);
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;
621 /* Get the size of the first reported zone */
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");
631 *zblocks = zone_blocks;
636 static void sd_zbc_print_zones(struct scsi_disk *sdkp)
638 if (!sd_is_zoned(sdkp) || !sdkp->capacity)
641 if (sdkp->capacity & (sdkp->zone_blocks - 1))
642 sd_printk(KERN_NOTICE, sdkp,
643 "%u zones of %u logical blocks + 1 runt zone\n",
647 sd_printk(KERN_NOTICE, sdkp,
648 "%u zones of %u logical blocks\n",
653 static void sd_zbc_revalidate_zones_cb(struct gendisk *disk)
655 struct scsi_disk *sdkp = scsi_disk(disk);
657 swap(sdkp->zones_wp_offset, sdkp->rev_wp_offset);
660 int sd_zbc_revalidate_zones(struct scsi_disk *sdkp)
662 struct gendisk *disk = sdkp->disk;
663 struct request_queue *q = disk->queue;
664 u32 zone_blocks = sdkp->rev_zone_blocks;
665 unsigned int nr_zones = sdkp->rev_nr_zones;
669 if (!sd_is_zoned(sdkp))
673 * Make sure revalidate zones are serialized to ensure exclusive
674 * updates of the scsi disk data.
676 mutex_lock(&sdkp->rev_mutex);
678 if (sdkp->zone_blocks == zone_blocks &&
679 sdkp->nr_zones == nr_zones &&
680 disk->queue->nr_zones == nr_zones)
683 sdkp->zone_blocks = zone_blocks;
684 sdkp->nr_zones = nr_zones;
685 sdkp->rev_wp_offset = kvcalloc(nr_zones, sizeof(u32), GFP_NOIO);
686 if (!sdkp->rev_wp_offset) {
691 ret = blk_revalidate_disk_zones(disk, sd_zbc_revalidate_zones_cb);
693 kvfree(sdkp->rev_wp_offset);
694 sdkp->rev_wp_offset = NULL;
697 sdkp->zone_blocks = 0;
703 max_append = min_t(u32, logical_to_sectors(sdkp->device, zone_blocks),
704 q->limits.max_segments << (PAGE_SHIFT - 9));
705 max_append = min_t(u32, max_append, queue_max_hw_sectors(q));
707 blk_queue_max_zone_append_sectors(q, max_append);
709 sd_zbc_print_zones(sdkp);
712 mutex_unlock(&sdkp->rev_mutex);
717 int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
719 struct gendisk *disk = sdkp->disk;
720 struct request_queue *q = disk->queue;
721 unsigned int nr_zones;
725 if (!sd_is_zoned(sdkp))
727 * Device managed or normal SCSI disk,
728 * no special handling required
732 /* Check zoned block device characteristics (unconstrained reads) */
733 ret = sd_zbc_check_zoned_characteristics(sdkp, buf);
737 /* Check the device capacity reported by report zones */
738 ret = sd_zbc_check_capacity(sdkp, buf, &zone_blocks);
742 /* The drive satisfies the kernel restrictions: set it up */
743 blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
744 blk_queue_required_elevator_features(q, ELEVATOR_F_ZBD_SEQ_WRITE);
745 nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
747 /* READ16/WRITE16 is mandatory for ZBC disks */
748 sdkp->device->use_16_for_rw = 1;
749 sdkp->device->use_10_for_rw = 0;
751 sdkp->rev_nr_zones = nr_zones;
752 sdkp->rev_zone_blocks = zone_blocks;
762 int sd_zbc_init_disk(struct scsi_disk *sdkp)
764 if (!sd_is_zoned(sdkp))
767 sdkp->zones_wp_offset = NULL;
768 spin_lock_init(&sdkp->zones_wp_offset_lock);
769 sdkp->rev_wp_offset = NULL;
770 mutex_init(&sdkp->rev_mutex);
771 INIT_WORK(&sdkp->zone_wp_offset_work, sd_zbc_update_wp_offset_workfn);
772 sdkp->zone_wp_update_buf = kzalloc(SD_BUF_SIZE, GFP_KERNEL);
773 if (!sdkp->zone_wp_update_buf)
779 void sd_zbc_release_disk(struct scsi_disk *sdkp)
781 kvfree(sdkp->zones_wp_offset);
782 sdkp->zones_wp_offset = NULL;
783 kfree(sdkp->zone_wp_update_buf);
784 sdkp->zone_wp_update_buf = NULL;