Loading...
Note: File does not exist in v3.15.
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#define CREATE_TRACE_POINTS
24#include "sd_trace.h"
25
26/**
27 * sd_zbc_get_zone_wp_offset - Get zone write pointer offset.
28 * @zone: Zone for which to return the write pointer offset.
29 *
30 * Return: offset of the write pointer from the start of the zone.
31 */
32static unsigned int sd_zbc_get_zone_wp_offset(struct blk_zone *zone)
33{
34 if (zone->type == ZBC_ZONE_TYPE_CONV)
35 return 0;
36
37 switch (zone->cond) {
38 case BLK_ZONE_COND_IMP_OPEN:
39 case BLK_ZONE_COND_EXP_OPEN:
40 case BLK_ZONE_COND_CLOSED:
41 return zone->wp - zone->start;
42 case BLK_ZONE_COND_FULL:
43 return zone->len;
44 case BLK_ZONE_COND_EMPTY:
45 case BLK_ZONE_COND_OFFLINE:
46 case BLK_ZONE_COND_READONLY:
47 default:
48 /*
49 * Offline and read-only zones do not have a valid
50 * write pointer. Use 0 as for an empty zone.
51 */
52 return 0;
53 }
54}
55
56/* Whether or not a SCSI zone descriptor describes a gap zone. */
57static bool sd_zbc_is_gap_zone(const u8 buf[64])
58{
59 return (buf[0] & 0xf) == ZBC_ZONE_TYPE_GAP;
60}
61
62/**
63 * sd_zbc_parse_report - Parse a SCSI zone descriptor
64 * @sdkp: SCSI disk pointer.
65 * @buf: SCSI zone descriptor.
66 * @idx: Index of the zone relative to the first zone reported by the current
67 * sd_zbc_report_zones() call.
68 * @cb: Callback function pointer.
69 * @data: Second argument passed to @cb.
70 *
71 * Return: Value returned by @cb.
72 *
73 * Convert a SCSI zone descriptor into struct blk_zone format. Additionally,
74 * call @cb(blk_zone, @data).
75 */
76static int sd_zbc_parse_report(struct scsi_disk *sdkp, const u8 buf[64],
77 unsigned int idx, report_zones_cb cb, void *data)
78{
79 struct scsi_device *sdp = sdkp->device;
80 struct blk_zone zone = { 0 };
81 sector_t start_lba, gran;
82 int ret;
83
84 if (WARN_ON_ONCE(sd_zbc_is_gap_zone(buf)))
85 return -EINVAL;
86
87 zone.type = buf[0] & 0x0f;
88 zone.cond = (buf[1] >> 4) & 0xf;
89 if (buf[1] & 0x01)
90 zone.reset = 1;
91 if (buf[1] & 0x02)
92 zone.non_seq = 1;
93
94 start_lba = get_unaligned_be64(&buf[16]);
95 zone.start = logical_to_sectors(sdp, start_lba);
96 zone.capacity = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
97 zone.len = zone.capacity;
98 if (sdkp->zone_starting_lba_gran) {
99 gran = logical_to_sectors(sdp, sdkp->zone_starting_lba_gran);
100 if (zone.len > gran) {
101 sd_printk(KERN_ERR, sdkp,
102 "Invalid zone at LBA %llu with capacity %llu and length %llu; granularity = %llu\n",
103 start_lba,
104 sectors_to_logical(sdp, zone.capacity),
105 sectors_to_logical(sdp, zone.len),
106 sectors_to_logical(sdp, gran));
107 return -EINVAL;
108 }
109 /*
110 * Use the starting LBA granularity instead of the zone length
111 * obtained from the REPORT ZONES command.
112 */
113 zone.len = gran;
114 }
115 if (zone.cond == ZBC_ZONE_COND_FULL)
116 zone.wp = zone.start + zone.len;
117 else
118 zone.wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24]));
119
120 ret = cb(&zone, idx, data);
121 if (ret)
122 return ret;
123
124 if (sdkp->rev_wp_offset)
125 sdkp->rev_wp_offset[idx] = sd_zbc_get_zone_wp_offset(&zone);
126
127 return 0;
128}
129
130/**
131 * sd_zbc_do_report_zones - Issue a REPORT ZONES scsi command.
132 * @sdkp: The target disk
133 * @buf: vmalloc-ed buffer to use for the reply
134 * @buflen: the buffer size
135 * @lba: Start LBA of the report
136 * @partial: Do partial report
137 *
138 * For internal use during device validation.
139 * Using partial=true can significantly speed up execution of a report zones
140 * command because the disk does not have to count all possible report matching
141 * zones and will only report the count of zones fitting in the command reply
142 * buffer.
143 */
144static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
145 unsigned int buflen, sector_t lba,
146 bool partial)
147{
148 struct scsi_device *sdp = sdkp->device;
149 const int timeout = sdp->request_queue->rq_timeout;
150 struct scsi_sense_hdr sshdr;
151 unsigned char cmd[16];
152 unsigned int rep_len;
153 int result;
154
155 memset(cmd, 0, 16);
156 cmd[0] = ZBC_IN;
157 cmd[1] = ZI_REPORT_ZONES;
158 put_unaligned_be64(lba, &cmd[2]);
159 put_unaligned_be32(buflen, &cmd[10]);
160 if (partial)
161 cmd[14] = ZBC_REPORT_ZONE_PARTIAL;
162
163 result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
164 buf, buflen, &sshdr,
165 timeout, SD_MAX_RETRIES, NULL);
166 if (result) {
167 sd_printk(KERN_ERR, sdkp,
168 "REPORT ZONES start lba %llu failed\n", lba);
169 sd_print_result(sdkp, "REPORT ZONES", result);
170 if (result > 0 && scsi_sense_valid(&sshdr))
171 sd_print_sense_hdr(sdkp, &sshdr);
172 return -EIO;
173 }
174
175 rep_len = get_unaligned_be32(&buf[0]);
176 if (rep_len < 64) {
177 sd_printk(KERN_ERR, sdkp,
178 "REPORT ZONES report invalid length %u\n",
179 rep_len);
180 return -EIO;
181 }
182
183 return 0;
184}
185
186/**
187 * sd_zbc_alloc_report_buffer() - Allocate a buffer for report zones reply.
188 * @sdkp: The target disk
189 * @nr_zones: Maximum number of zones to report
190 * @buflen: Size of the buffer allocated
191 *
192 * Try to allocate a reply buffer for the number of requested zones.
193 * The size of the buffer allocated may be smaller than requested to
194 * satify the device constraint (max_hw_sectors, max_segments, etc).
195 *
196 * Return the address of the allocated buffer and update @buflen with
197 * the size of the allocated buffer.
198 */
199static void *sd_zbc_alloc_report_buffer(struct scsi_disk *sdkp,
200 unsigned int nr_zones, size_t *buflen)
201{
202 struct request_queue *q = sdkp->disk->queue;
203 size_t bufsize;
204 void *buf;
205
206 /*
207 * Report zone buffer size should be at most 64B times the number of
208 * zones requested plus the 64B reply header, but should be aligned
209 * to SECTOR_SIZE for ATA devices.
210 * Make sure that this size does not exceed the hardware capabilities.
211 * Furthermore, since the report zone command cannot be split, make
212 * sure that the allocated buffer can always be mapped by limiting the
213 * number of pages allocated to the HBA max segments limit.
214 */
215 nr_zones = min(nr_zones, sdkp->zone_info.nr_zones);
216 bufsize = roundup((nr_zones + 1) * 64, SECTOR_SIZE);
217 bufsize = min_t(size_t, bufsize,
218 queue_max_hw_sectors(q) << SECTOR_SHIFT);
219 bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
220
221 while (bufsize >= SECTOR_SIZE) {
222 buf = __vmalloc(bufsize,
223 GFP_KERNEL | __GFP_ZERO | __GFP_NORETRY);
224 if (buf) {
225 *buflen = bufsize;
226 return buf;
227 }
228 bufsize = rounddown(bufsize >> 1, SECTOR_SIZE);
229 }
230
231 return NULL;
232}
233
234/**
235 * sd_zbc_zone_sectors - Get the device zone size in number of 512B sectors.
236 * @sdkp: The target disk
237 */
238static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
239{
240 return logical_to_sectors(sdkp->device, sdkp->zone_info.zone_blocks);
241}
242
243/**
244 * sd_zbc_report_zones - SCSI .report_zones() callback.
245 * @disk: Disk to report zones for.
246 * @sector: Start sector.
247 * @nr_zones: Maximum number of zones to report.
248 * @cb: Callback function called to report zone information.
249 * @data: Second argument passed to @cb.
250 *
251 * Called by the block layer to iterate over zone information. See also the
252 * disk->fops->report_zones() calls in block/blk-zoned.c.
253 */
254int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
255 unsigned int nr_zones, report_zones_cb cb, void *data)
256{
257 struct scsi_disk *sdkp = scsi_disk(disk);
258 sector_t lba = sectors_to_logical(sdkp->device, sector);
259 unsigned int nr, i;
260 unsigned char *buf;
261 u64 zone_length, start_lba;
262 size_t offset, buflen = 0;
263 int zone_idx = 0;
264 int ret;
265
266 if (!sd_is_zoned(sdkp))
267 /* Not a zoned device */
268 return -EOPNOTSUPP;
269
270 if (!sdkp->capacity)
271 /* Device gone or invalid */
272 return -ENODEV;
273
274 buf = sd_zbc_alloc_report_buffer(sdkp, nr_zones, &buflen);
275 if (!buf)
276 return -ENOMEM;
277
278 while (zone_idx < nr_zones && lba < sdkp->capacity) {
279 ret = sd_zbc_do_report_zones(sdkp, buf, buflen, lba, true);
280 if (ret)
281 goto out;
282
283 offset = 0;
284 nr = min(nr_zones, get_unaligned_be32(&buf[0]) / 64);
285 if (!nr)
286 break;
287
288 for (i = 0; i < nr && zone_idx < nr_zones; i++) {
289 offset += 64;
290 start_lba = get_unaligned_be64(&buf[offset + 16]);
291 zone_length = get_unaligned_be64(&buf[offset + 8]);
292 if ((zone_idx == 0 &&
293 (lba < start_lba ||
294 lba >= start_lba + zone_length)) ||
295 (zone_idx > 0 && start_lba != lba) ||
296 start_lba + zone_length < start_lba) {
297 sd_printk(KERN_ERR, sdkp,
298 "Zone %d at LBA %llu is invalid: %llu + %llu\n",
299 zone_idx, lba, start_lba, zone_length);
300 ret = -EINVAL;
301 goto out;
302 }
303 lba = start_lba + zone_length;
304 if (sd_zbc_is_gap_zone(&buf[offset])) {
305 if (sdkp->zone_starting_lba_gran)
306 continue;
307 sd_printk(KERN_ERR, sdkp,
308 "Gap zone without constant LBA offsets\n");
309 ret = -EINVAL;
310 goto out;
311 }
312
313 ret = sd_zbc_parse_report(sdkp, buf + offset, zone_idx,
314 cb, data);
315 if (ret)
316 goto out;
317
318 zone_idx++;
319 }
320 }
321
322 ret = zone_idx;
323out:
324 kvfree(buf);
325 return ret;
326}
327
328static blk_status_t sd_zbc_cmnd_checks(struct scsi_cmnd *cmd)
329{
330 struct request *rq = scsi_cmd_to_rq(cmd);
331 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
332 sector_t sector = blk_rq_pos(rq);
333
334 if (!sd_is_zoned(sdkp))
335 /* Not a zoned device */
336 return BLK_STS_IOERR;
337
338 if (sdkp->device->changed)
339 return BLK_STS_IOERR;
340
341 if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
342 /* Unaligned request */
343 return BLK_STS_IOERR;
344
345 return BLK_STS_OK;
346}
347
348#define SD_ZBC_INVALID_WP_OFST (~0u)
349#define SD_ZBC_UPDATING_WP_OFST (SD_ZBC_INVALID_WP_OFST - 1)
350
351static int sd_zbc_update_wp_offset_cb(struct blk_zone *zone, unsigned int idx,
352 void *data)
353{
354 struct scsi_disk *sdkp = data;
355
356 lockdep_assert_held(&sdkp->zones_wp_offset_lock);
357
358 sdkp->zones_wp_offset[idx] = sd_zbc_get_zone_wp_offset(zone);
359
360 return 0;
361}
362
363/*
364 * An attempt to append a zone triggered an invalid write pointer error.
365 * Reread the write pointer of the zone(s) in which the append failed.
366 */
367static void sd_zbc_update_wp_offset_workfn(struct work_struct *work)
368{
369 struct scsi_disk *sdkp;
370 unsigned long flags;
371 sector_t zno;
372 int ret;
373
374 sdkp = container_of(work, struct scsi_disk, zone_wp_offset_work);
375
376 spin_lock_irqsave(&sdkp->zones_wp_offset_lock, flags);
377 for (zno = 0; zno < sdkp->zone_info.nr_zones; zno++) {
378 if (sdkp->zones_wp_offset[zno] != SD_ZBC_UPDATING_WP_OFST)
379 continue;
380
381 spin_unlock_irqrestore(&sdkp->zones_wp_offset_lock, flags);
382 ret = sd_zbc_do_report_zones(sdkp, sdkp->zone_wp_update_buf,
383 SD_BUF_SIZE,
384 zno * sdkp->zone_info.zone_blocks, true);
385 spin_lock_irqsave(&sdkp->zones_wp_offset_lock, flags);
386 if (!ret)
387 sd_zbc_parse_report(sdkp, sdkp->zone_wp_update_buf + 64,
388 zno, sd_zbc_update_wp_offset_cb,
389 sdkp);
390 }
391 spin_unlock_irqrestore(&sdkp->zones_wp_offset_lock, flags);
392
393 scsi_device_put(sdkp->device);
394}
395
396/**
397 * sd_zbc_prepare_zone_append() - Prepare an emulated ZONE_APPEND command.
398 * @cmd: the command to setup
399 * @lba: the LBA to patch
400 * @nr_blocks: the number of LBAs to be written
401 *
402 * Called from sd_setup_read_write_cmnd() for REQ_OP_ZONE_APPEND.
403 * @sd_zbc_prepare_zone_append() handles the necessary zone wrote locking and
404 * patching of the lba for an emulated ZONE_APPEND command.
405 *
406 * In case the cached write pointer offset is %SD_ZBC_INVALID_WP_OFST it will
407 * schedule a REPORT ZONES command and return BLK_STS_IOERR.
408 */
409blk_status_t sd_zbc_prepare_zone_append(struct scsi_cmnd *cmd, sector_t *lba,
410 unsigned int nr_blocks)
411{
412 struct request *rq = scsi_cmd_to_rq(cmd);
413 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
414 unsigned int wp_offset, zno = blk_rq_zone_no(rq);
415 unsigned long flags;
416 blk_status_t ret;
417
418 ret = sd_zbc_cmnd_checks(cmd);
419 if (ret != BLK_STS_OK)
420 return ret;
421
422 if (!blk_rq_zone_is_seq(rq))
423 return BLK_STS_IOERR;
424
425 /* Unlock of the write lock will happen in sd_zbc_complete() */
426 if (!blk_req_zone_write_trylock(rq))
427 return BLK_STS_ZONE_RESOURCE;
428
429 spin_lock_irqsave(&sdkp->zones_wp_offset_lock, flags);
430 wp_offset = sdkp->zones_wp_offset[zno];
431 switch (wp_offset) {
432 case SD_ZBC_INVALID_WP_OFST:
433 /*
434 * We are about to schedule work to update a zone write pointer
435 * offset, which will cause the zone append command to be
436 * requeued. So make sure that the scsi device does not go away
437 * while the work is being processed.
438 */
439 if (scsi_device_get(sdkp->device)) {
440 ret = BLK_STS_IOERR;
441 break;
442 }
443 sdkp->zones_wp_offset[zno] = SD_ZBC_UPDATING_WP_OFST;
444 schedule_work(&sdkp->zone_wp_offset_work);
445 fallthrough;
446 case SD_ZBC_UPDATING_WP_OFST:
447 ret = BLK_STS_DEV_RESOURCE;
448 break;
449 default:
450 wp_offset = sectors_to_logical(sdkp->device, wp_offset);
451 if (wp_offset + nr_blocks > sdkp->zone_info.zone_blocks) {
452 ret = BLK_STS_IOERR;
453 break;
454 }
455
456 trace_scsi_prepare_zone_append(cmd, *lba, wp_offset);
457 *lba += wp_offset;
458 }
459 spin_unlock_irqrestore(&sdkp->zones_wp_offset_lock, flags);
460 if (ret)
461 blk_req_zone_write_unlock(rq);
462 return ret;
463}
464
465/**
466 * sd_zbc_setup_zone_mgmt_cmnd - Prepare a zone ZBC_OUT command. The operations
467 * can be RESET WRITE POINTER, OPEN, CLOSE or FINISH.
468 * @cmd: the command to setup
469 * @op: Operation to be performed
470 * @all: All zones control
471 *
472 * Called from sd_init_command() for REQ_OP_ZONE_RESET, REQ_OP_ZONE_RESET_ALL,
473 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE or REQ_OP_ZONE_FINISH requests.
474 */
475blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
476 unsigned char op, bool all)
477{
478 struct request *rq = scsi_cmd_to_rq(cmd);
479 sector_t sector = blk_rq_pos(rq);
480 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
481 sector_t block = sectors_to_logical(sdkp->device, sector);
482 blk_status_t ret;
483
484 ret = sd_zbc_cmnd_checks(cmd);
485 if (ret != BLK_STS_OK)
486 return ret;
487
488 cmd->cmd_len = 16;
489 memset(cmd->cmnd, 0, cmd->cmd_len);
490 cmd->cmnd[0] = ZBC_OUT;
491 cmd->cmnd[1] = op;
492 if (all)
493 cmd->cmnd[14] = 0x1;
494 else
495 put_unaligned_be64(block, &cmd->cmnd[2]);
496
497 rq->timeout = SD_TIMEOUT;
498 cmd->sc_data_direction = DMA_NONE;
499 cmd->transfersize = 0;
500 cmd->allowed = 0;
501
502 return BLK_STS_OK;
503}
504
505static bool sd_zbc_need_zone_wp_update(struct request *rq)
506{
507 switch (req_op(rq)) {
508 case REQ_OP_ZONE_APPEND:
509 case REQ_OP_ZONE_FINISH:
510 case REQ_OP_ZONE_RESET:
511 case REQ_OP_ZONE_RESET_ALL:
512 return true;
513 case REQ_OP_WRITE:
514 case REQ_OP_WRITE_ZEROES:
515 return blk_rq_zone_is_seq(rq);
516 default:
517 return false;
518 }
519}
520
521/**
522 * sd_zbc_zone_wp_update - Update cached zone write pointer upon cmd completion
523 * @cmd: Completed command
524 * @good_bytes: Command reply bytes
525 *
526 * Called from sd_zbc_complete() to handle the update of the cached zone write
527 * pointer value in case an update is needed.
528 */
529static unsigned int sd_zbc_zone_wp_update(struct scsi_cmnd *cmd,
530 unsigned int good_bytes)
531{
532 int result = cmd->result;
533 struct request *rq = scsi_cmd_to_rq(cmd);
534 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
535 unsigned int zno = blk_rq_zone_no(rq);
536 enum req_op op = req_op(rq);
537 unsigned long flags;
538
539 /*
540 * If we got an error for a command that needs updating the write
541 * pointer offset cache, we must mark the zone wp offset entry as
542 * invalid to force an update from disk the next time a zone append
543 * command is issued.
544 */
545 spin_lock_irqsave(&sdkp->zones_wp_offset_lock, flags);
546
547 if (result && op != REQ_OP_ZONE_RESET_ALL) {
548 if (op == REQ_OP_ZONE_APPEND) {
549 /* Force complete completion (no retry) */
550 good_bytes = 0;
551 scsi_set_resid(cmd, blk_rq_bytes(rq));
552 }
553
554 /*
555 * Force an update of the zone write pointer offset on
556 * the next zone append access.
557 */
558 if (sdkp->zones_wp_offset[zno] != SD_ZBC_UPDATING_WP_OFST)
559 sdkp->zones_wp_offset[zno] = SD_ZBC_INVALID_WP_OFST;
560 goto unlock_wp_offset;
561 }
562
563 switch (op) {
564 case REQ_OP_ZONE_APPEND:
565 trace_scsi_zone_wp_update(cmd, rq->__sector,
566 sdkp->zones_wp_offset[zno], good_bytes);
567 rq->__sector += sdkp->zones_wp_offset[zno];
568 fallthrough;
569 case REQ_OP_WRITE_ZEROES:
570 case REQ_OP_WRITE:
571 if (sdkp->zones_wp_offset[zno] < sd_zbc_zone_sectors(sdkp))
572 sdkp->zones_wp_offset[zno] +=
573 good_bytes >> SECTOR_SHIFT;
574 break;
575 case REQ_OP_ZONE_RESET:
576 sdkp->zones_wp_offset[zno] = 0;
577 break;
578 case REQ_OP_ZONE_FINISH:
579 sdkp->zones_wp_offset[zno] = sd_zbc_zone_sectors(sdkp);
580 break;
581 case REQ_OP_ZONE_RESET_ALL:
582 memset(sdkp->zones_wp_offset, 0,
583 sdkp->zone_info.nr_zones * sizeof(unsigned int));
584 break;
585 default:
586 break;
587 }
588
589unlock_wp_offset:
590 spin_unlock_irqrestore(&sdkp->zones_wp_offset_lock, flags);
591
592 return good_bytes;
593}
594
595/**
596 * sd_zbc_complete - ZBC command post processing.
597 * @cmd: Completed command
598 * @good_bytes: Command reply bytes
599 * @sshdr: command sense header
600 *
601 * Called from sd_done() to handle zone commands errors and updates to the
602 * device queue zone write pointer offset cahce.
603 */
604unsigned int sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
605 struct scsi_sense_hdr *sshdr)
606{
607 int result = cmd->result;
608 struct request *rq = scsi_cmd_to_rq(cmd);
609
610 if (op_is_zone_mgmt(req_op(rq)) &&
611 result &&
612 sshdr->sense_key == ILLEGAL_REQUEST &&
613 sshdr->asc == 0x24) {
614 /*
615 * INVALID FIELD IN CDB error: a zone management command was
616 * attempted on a conventional zone. Nothing to worry about,
617 * so be quiet about the error.
618 */
619 rq->rq_flags |= RQF_QUIET;
620 } else if (sd_zbc_need_zone_wp_update(rq))
621 good_bytes = sd_zbc_zone_wp_update(cmd, good_bytes);
622
623 if (req_op(rq) == REQ_OP_ZONE_APPEND)
624 blk_req_zone_write_unlock(rq);
625
626 return good_bytes;
627}
628
629/**
630 * sd_zbc_check_zoned_characteristics - Check zoned block device characteristics
631 * @sdkp: Target disk
632 * @buf: Buffer where to store the VPD page data
633 *
634 * Read VPD page B6, get information and check that reads are unconstrained.
635 */
636static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
637 unsigned char *buf)
638{
639 u64 zone_starting_lba_gran;
640
641 if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
642 sd_printk(KERN_NOTICE, sdkp,
643 "Read zoned characteristics VPD page failed\n");
644 return -ENODEV;
645 }
646
647 if (sdkp->device->type != TYPE_ZBC) {
648 /* Host-aware */
649 sdkp->urswrz = 1;
650 sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]);
651 sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]);
652 sdkp->zones_max_open = 0;
653 return 0;
654 }
655
656 /* Host-managed */
657 sdkp->urswrz = buf[4] & 1;
658 sdkp->zones_optimal_open = 0;
659 sdkp->zones_optimal_nonseq = 0;
660 sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
661 /* Check zone alignment method */
662 switch (buf[23] & 0xf) {
663 case 0:
664 case ZBC_CONSTANT_ZONE_LENGTH:
665 /* Use zone length */
666 break;
667 case ZBC_CONSTANT_ZONE_START_OFFSET:
668 zone_starting_lba_gran = get_unaligned_be64(&buf[24]);
669 if (zone_starting_lba_gran == 0 ||
670 !is_power_of_2(zone_starting_lba_gran) ||
671 logical_to_sectors(sdkp->device, zone_starting_lba_gran) >
672 UINT_MAX) {
673 sd_printk(KERN_ERR, sdkp,
674 "Invalid zone starting LBA granularity %llu\n",
675 zone_starting_lba_gran);
676 return -ENODEV;
677 }
678 sdkp->zone_starting_lba_gran = zone_starting_lba_gran;
679 break;
680 default:
681 sd_printk(KERN_ERR, sdkp, "Invalid zone alignment method\n");
682 return -ENODEV;
683 }
684
685 /*
686 * Check for unconstrained reads: host-managed devices with
687 * constrained reads (drives failing read after write pointer)
688 * are not supported.
689 */
690 if (!sdkp->urswrz) {
691 if (sdkp->first_scan)
692 sd_printk(KERN_NOTICE, sdkp,
693 "constrained reads devices are not supported\n");
694 return -ENODEV;
695 }
696
697 return 0;
698}
699
700/**
701 * sd_zbc_check_capacity - Check the device capacity
702 * @sdkp: Target disk
703 * @buf: command buffer
704 * @zblocks: zone size in logical blocks
705 *
706 * Get the device zone size and check that the device capacity as reported
707 * by READ CAPACITY matches the max_lba value (plus one) of the report zones
708 * command reply for devices with RC_BASIS == 0.
709 *
710 * Returns 0 upon success or an error code upon failure.
711 */
712static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf,
713 u32 *zblocks)
714{
715 u64 zone_blocks;
716 sector_t max_lba;
717 unsigned char *rec;
718 int ret;
719
720 /* Do a report zone to get max_lba and the size of the first zone */
721 ret = sd_zbc_do_report_zones(sdkp, buf, SD_BUF_SIZE, 0, false);
722 if (ret)
723 return ret;
724
725 if (sdkp->rc_basis == 0) {
726 /* The max_lba field is the capacity of this device */
727 max_lba = get_unaligned_be64(&buf[8]);
728 if (sdkp->capacity != max_lba + 1) {
729 if (sdkp->first_scan)
730 sd_printk(KERN_WARNING, sdkp,
731 "Changing capacity from %llu to max LBA+1 %llu\n",
732 (unsigned long long)sdkp->capacity,
733 (unsigned long long)max_lba + 1);
734 sdkp->capacity = max_lba + 1;
735 }
736 }
737
738 if (sdkp->zone_starting_lba_gran == 0) {
739 /* Get the size of the first reported zone */
740 rec = buf + 64;
741 zone_blocks = get_unaligned_be64(&rec[8]);
742 if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
743 if (sdkp->first_scan)
744 sd_printk(KERN_NOTICE, sdkp,
745 "Zone size too large\n");
746 return -EFBIG;
747 }
748 } else {
749 zone_blocks = sdkp->zone_starting_lba_gran;
750 }
751
752 if (!is_power_of_2(zone_blocks)) {
753 sd_printk(KERN_ERR, sdkp,
754 "Zone size %llu is not a power of two.\n",
755 zone_blocks);
756 return -EINVAL;
757 }
758
759 *zblocks = zone_blocks;
760
761 return 0;
762}
763
764static void sd_zbc_print_zones(struct scsi_disk *sdkp)
765{
766 if (!sd_is_zoned(sdkp) || !sdkp->capacity)
767 return;
768
769 if (sdkp->capacity & (sdkp->zone_info.zone_blocks - 1))
770 sd_printk(KERN_NOTICE, sdkp,
771 "%u zones of %u logical blocks + 1 runt zone\n",
772 sdkp->zone_info.nr_zones - 1,
773 sdkp->zone_info.zone_blocks);
774 else
775 sd_printk(KERN_NOTICE, sdkp,
776 "%u zones of %u logical blocks\n",
777 sdkp->zone_info.nr_zones,
778 sdkp->zone_info.zone_blocks);
779}
780
781static int sd_zbc_init_disk(struct scsi_disk *sdkp)
782{
783 sdkp->zones_wp_offset = NULL;
784 spin_lock_init(&sdkp->zones_wp_offset_lock);
785 sdkp->rev_wp_offset = NULL;
786 mutex_init(&sdkp->rev_mutex);
787 INIT_WORK(&sdkp->zone_wp_offset_work, sd_zbc_update_wp_offset_workfn);
788 sdkp->zone_wp_update_buf = kzalloc(SD_BUF_SIZE, GFP_KERNEL);
789 if (!sdkp->zone_wp_update_buf)
790 return -ENOMEM;
791
792 return 0;
793}
794
795void sd_zbc_free_zone_info(struct scsi_disk *sdkp)
796{
797 if (!sdkp->zone_wp_update_buf)
798 return;
799
800 /* Serialize against revalidate zones */
801 mutex_lock(&sdkp->rev_mutex);
802
803 kvfree(sdkp->zones_wp_offset);
804 sdkp->zones_wp_offset = NULL;
805 kfree(sdkp->zone_wp_update_buf);
806 sdkp->zone_wp_update_buf = NULL;
807
808 sdkp->early_zone_info = (struct zoned_disk_info){ };
809 sdkp->zone_info = (struct zoned_disk_info){ };
810
811 mutex_unlock(&sdkp->rev_mutex);
812}
813
814static void sd_zbc_revalidate_zones_cb(struct gendisk *disk)
815{
816 struct scsi_disk *sdkp = scsi_disk(disk);
817
818 swap(sdkp->zones_wp_offset, sdkp->rev_wp_offset);
819}
820
821/*
822 * Call blk_revalidate_disk_zones() if any of the zoned disk properties have
823 * changed that make it necessary to call that function. Called by
824 * sd_revalidate_disk() after the gendisk capacity has been set.
825 */
826int sd_zbc_revalidate_zones(struct scsi_disk *sdkp)
827{
828 struct gendisk *disk = sdkp->disk;
829 struct request_queue *q = disk->queue;
830 u32 zone_blocks = sdkp->early_zone_info.zone_blocks;
831 unsigned int nr_zones = sdkp->early_zone_info.nr_zones;
832 u32 max_append;
833 int ret = 0;
834 unsigned int flags;
835
836 /*
837 * For all zoned disks, initialize zone append emulation data if not
838 * already done. This is necessary also for host-aware disks used as
839 * regular disks due to the presence of partitions as these partitions
840 * may be deleted and the disk zoned model changed back from
841 * BLK_ZONED_NONE to BLK_ZONED_HA.
842 */
843 if (sd_is_zoned(sdkp) && !sdkp->zone_wp_update_buf) {
844 ret = sd_zbc_init_disk(sdkp);
845 if (ret)
846 return ret;
847 }
848
849 /*
850 * There is nothing to do for regular disks, including host-aware disks
851 * that have partitions.
852 */
853 if (!blk_queue_is_zoned(q))
854 return 0;
855
856 /*
857 * Make sure revalidate zones are serialized to ensure exclusive
858 * updates of the scsi disk data.
859 */
860 mutex_lock(&sdkp->rev_mutex);
861
862 if (sdkp->zone_info.zone_blocks == zone_blocks &&
863 sdkp->zone_info.nr_zones == nr_zones &&
864 disk->nr_zones == nr_zones)
865 goto unlock;
866
867 flags = memalloc_noio_save();
868 sdkp->zone_info.zone_blocks = zone_blocks;
869 sdkp->zone_info.nr_zones = nr_zones;
870 sdkp->rev_wp_offset = kvcalloc(nr_zones, sizeof(u32), GFP_KERNEL);
871 if (!sdkp->rev_wp_offset) {
872 ret = -ENOMEM;
873 memalloc_noio_restore(flags);
874 goto unlock;
875 }
876
877 ret = blk_revalidate_disk_zones(disk, sd_zbc_revalidate_zones_cb);
878
879 memalloc_noio_restore(flags);
880 kvfree(sdkp->rev_wp_offset);
881 sdkp->rev_wp_offset = NULL;
882
883 if (ret) {
884 sdkp->zone_info = (struct zoned_disk_info){ };
885 sdkp->capacity = 0;
886 goto unlock;
887 }
888
889 max_append = min_t(u32, logical_to_sectors(sdkp->device, zone_blocks),
890 q->limits.max_segments << (PAGE_SHIFT - 9));
891 max_append = min_t(u32, max_append, queue_max_hw_sectors(q));
892
893 blk_queue_max_zone_append_sectors(q, max_append);
894
895 sd_zbc_print_zones(sdkp);
896
897unlock:
898 mutex_unlock(&sdkp->rev_mutex);
899
900 return ret;
901}
902
903/**
904 * sd_zbc_read_zones - Read zone information and update the request queue
905 * @sdkp: SCSI disk pointer.
906 * @buf: 512 byte buffer used for storing SCSI command output.
907 *
908 * Read zone information and update the request queue zone characteristics and
909 * also the zoned device information in *sdkp. Called by sd_revalidate_disk()
910 * before the gendisk capacity has been set.
911 */
912int sd_zbc_read_zones(struct scsi_disk *sdkp, u8 buf[SD_BUF_SIZE])
913{
914 struct gendisk *disk = sdkp->disk;
915 struct request_queue *q = disk->queue;
916 unsigned int nr_zones;
917 u32 zone_blocks = 0;
918 int ret;
919
920 if (!sd_is_zoned(sdkp)) {
921 /*
922 * Device managed or normal SCSI disk, no special handling
923 * required. Nevertheless, free the disk zone information in
924 * case the device type changed.
925 */
926 sd_zbc_free_zone_info(sdkp);
927 return 0;
928 }
929
930 /* READ16/WRITE16/SYNC16 is mandatory for ZBC devices */
931 sdkp->device->use_16_for_rw = 1;
932 sdkp->device->use_10_for_rw = 0;
933 sdkp->device->use_16_for_sync = 1;
934
935 if (!blk_queue_is_zoned(q)) {
936 /*
937 * This can happen for a host aware disk with partitions.
938 * The block device zone model was already cleared by
939 * disk_set_zoned(). Only free the scsi disk zone
940 * information and exit early.
941 */
942 sd_zbc_free_zone_info(sdkp);
943 return 0;
944 }
945
946 /* Check zoned block device characteristics (unconstrained reads) */
947 ret = sd_zbc_check_zoned_characteristics(sdkp, buf);
948 if (ret)
949 goto err;
950
951 /* Check the device capacity reported by report zones */
952 ret = sd_zbc_check_capacity(sdkp, buf, &zone_blocks);
953 if (ret != 0)
954 goto err;
955
956 /* The drive satisfies the kernel restrictions: set it up */
957 blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
958 blk_queue_required_elevator_features(q, ELEVATOR_F_ZBD_SEQ_WRITE);
959 if (sdkp->zones_max_open == U32_MAX)
960 disk_set_max_open_zones(disk, 0);
961 else
962 disk_set_max_open_zones(disk, sdkp->zones_max_open);
963 disk_set_max_active_zones(disk, 0);
964 nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
965
966 /*
967 * Per ZBC and ZAC specifications, writes in sequential write required
968 * zones of host-managed devices must be aligned to the device physical
969 * block size.
970 */
971 if (blk_queue_zoned_model(q) == BLK_ZONED_HM)
972 blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
973
974 sdkp->early_zone_info.nr_zones = nr_zones;
975 sdkp->early_zone_info.zone_blocks = zone_blocks;
976
977 return 0;
978
979err:
980 sdkp->capacity = 0;
981
982 return ret;
983}