Loading...
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 const struct scsi_exec_args exec_args = {
152 .sshdr = &sshdr,
153 };
154 unsigned char cmd[16];
155 unsigned int rep_len;
156 int result;
157
158 memset(cmd, 0, 16);
159 cmd[0] = ZBC_IN;
160 cmd[1] = ZI_REPORT_ZONES;
161 put_unaligned_be64(lba, &cmd[2]);
162 put_unaligned_be32(buflen, &cmd[10]);
163 if (partial)
164 cmd[14] = ZBC_REPORT_ZONE_PARTIAL;
165
166 result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buf, buflen,
167 timeout, SD_MAX_RETRIES, &exec_args);
168 if (result) {
169 sd_printk(KERN_ERR, sdkp,
170 "REPORT ZONES start lba %llu failed\n", lba);
171 sd_print_result(sdkp, "REPORT ZONES", result);
172 if (result > 0 && scsi_sense_valid(&sshdr))
173 sd_print_sense_hdr(sdkp, &sshdr);
174 return -EIO;
175 }
176
177 rep_len = get_unaligned_be32(&buf[0]);
178 if (rep_len < 64) {
179 sd_printk(KERN_ERR, sdkp,
180 "REPORT ZONES report invalid length %u\n",
181 rep_len);
182 return -EIO;
183 }
184
185 return 0;
186}
187
188/**
189 * sd_zbc_alloc_report_buffer() - Allocate a buffer for report zones reply.
190 * @sdkp: The target disk
191 * @nr_zones: Maximum number of zones to report
192 * @buflen: Size of the buffer allocated
193 *
194 * Try to allocate a reply buffer for the number of requested zones.
195 * The size of the buffer allocated may be smaller than requested to
196 * satify the device constraint (max_hw_sectors, max_segments, etc).
197 *
198 * Return the address of the allocated buffer and update @buflen with
199 * the size of the allocated buffer.
200 */
201static void *sd_zbc_alloc_report_buffer(struct scsi_disk *sdkp,
202 unsigned int nr_zones, size_t *buflen)
203{
204 struct request_queue *q = sdkp->disk->queue;
205 size_t bufsize;
206 void *buf;
207
208 /*
209 * Report zone buffer size should be at most 64B times the number of
210 * zones requested plus the 64B reply header, but should be aligned
211 * to SECTOR_SIZE for ATA devices.
212 * Make sure that this size does not exceed the hardware capabilities.
213 * Furthermore, since the report zone command cannot be split, make
214 * sure that the allocated buffer can always be mapped by limiting the
215 * number of pages allocated to the HBA max segments limit.
216 */
217 nr_zones = min(nr_zones, sdkp->zone_info.nr_zones);
218 bufsize = roundup((nr_zones + 1) * 64, SECTOR_SIZE);
219 bufsize = min_t(size_t, bufsize,
220 queue_max_hw_sectors(q) << SECTOR_SHIFT);
221 bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
222
223 while (bufsize >= SECTOR_SIZE) {
224 buf = __vmalloc(bufsize,
225 GFP_KERNEL | __GFP_ZERO | __GFP_NORETRY);
226 if (buf) {
227 *buflen = bufsize;
228 return buf;
229 }
230 bufsize = rounddown(bufsize >> 1, SECTOR_SIZE);
231 }
232
233 return NULL;
234}
235
236/**
237 * sd_zbc_zone_sectors - Get the device zone size in number of 512B sectors.
238 * @sdkp: The target disk
239 */
240static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
241{
242 return logical_to_sectors(sdkp->device, sdkp->zone_info.zone_blocks);
243}
244
245/**
246 * sd_zbc_report_zones - SCSI .report_zones() callback.
247 * @disk: Disk to report zones for.
248 * @sector: Start sector.
249 * @nr_zones: Maximum number of zones to report.
250 * @cb: Callback function called to report zone information.
251 * @data: Second argument passed to @cb.
252 *
253 * Called by the block layer to iterate over zone information. See also the
254 * disk->fops->report_zones() calls in block/blk-zoned.c.
255 */
256int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
257 unsigned int nr_zones, report_zones_cb cb, void *data)
258{
259 struct scsi_disk *sdkp = scsi_disk(disk);
260 sector_t lba = sectors_to_logical(sdkp->device, sector);
261 unsigned int nr, i;
262 unsigned char *buf;
263 u64 zone_length, start_lba;
264 size_t offset, buflen = 0;
265 int zone_idx = 0;
266 int ret;
267
268 if (!sd_is_zoned(sdkp))
269 /* Not a zoned device */
270 return -EOPNOTSUPP;
271
272 if (!sdkp->capacity)
273 /* Device gone or invalid */
274 return -ENODEV;
275
276 buf = sd_zbc_alloc_report_buffer(sdkp, nr_zones, &buflen);
277 if (!buf)
278 return -ENOMEM;
279
280 while (zone_idx < nr_zones && lba < sdkp->capacity) {
281 ret = sd_zbc_do_report_zones(sdkp, buf, buflen, lba, true);
282 if (ret)
283 goto out;
284
285 offset = 0;
286 nr = min(nr_zones, get_unaligned_be32(&buf[0]) / 64);
287 if (!nr)
288 break;
289
290 for (i = 0; i < nr && zone_idx < nr_zones; i++) {
291 offset += 64;
292 start_lba = get_unaligned_be64(&buf[offset + 16]);
293 zone_length = get_unaligned_be64(&buf[offset + 8]);
294 if ((zone_idx == 0 &&
295 (lba < start_lba ||
296 lba >= start_lba + zone_length)) ||
297 (zone_idx > 0 && start_lba != lba) ||
298 start_lba + zone_length < start_lba) {
299 sd_printk(KERN_ERR, sdkp,
300 "Zone %d at LBA %llu is invalid: %llu + %llu\n",
301 zone_idx, lba, start_lba, zone_length);
302 ret = -EINVAL;
303 goto out;
304 }
305 lba = start_lba + zone_length;
306 if (sd_zbc_is_gap_zone(&buf[offset])) {
307 if (sdkp->zone_starting_lba_gran)
308 continue;
309 sd_printk(KERN_ERR, sdkp,
310 "Gap zone without constant LBA offsets\n");
311 ret = -EINVAL;
312 goto out;
313 }
314
315 ret = sd_zbc_parse_report(sdkp, buf + offset, zone_idx,
316 cb, data);
317 if (ret)
318 goto out;
319
320 zone_idx++;
321 }
322 }
323
324 ret = zone_idx;
325out:
326 kvfree(buf);
327 return ret;
328}
329
330static blk_status_t sd_zbc_cmnd_checks(struct scsi_cmnd *cmd)
331{
332 struct request *rq = scsi_cmd_to_rq(cmd);
333 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
334 sector_t sector = blk_rq_pos(rq);
335
336 if (!sd_is_zoned(sdkp))
337 /* Not a zoned device */
338 return BLK_STS_IOERR;
339
340 if (sdkp->device->changed)
341 return BLK_STS_IOERR;
342
343 if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
344 /* Unaligned request */
345 return BLK_STS_IOERR;
346
347 return BLK_STS_OK;
348}
349
350#define SD_ZBC_INVALID_WP_OFST (~0u)
351#define SD_ZBC_UPDATING_WP_OFST (SD_ZBC_INVALID_WP_OFST - 1)
352
353static int sd_zbc_update_wp_offset_cb(struct blk_zone *zone, unsigned int idx,
354 void *data)
355{
356 struct scsi_disk *sdkp = data;
357
358 lockdep_assert_held(&sdkp->zones_wp_offset_lock);
359
360 sdkp->zones_wp_offset[idx] = sd_zbc_get_zone_wp_offset(zone);
361
362 return 0;
363}
364
365/*
366 * An attempt to append a zone triggered an invalid write pointer error.
367 * Reread the write pointer of the zone(s) in which the append failed.
368 */
369static void sd_zbc_update_wp_offset_workfn(struct work_struct *work)
370{
371 struct scsi_disk *sdkp;
372 unsigned long flags;
373 sector_t zno;
374 int ret;
375
376 sdkp = container_of(work, struct scsi_disk, zone_wp_offset_work);
377
378 spin_lock_irqsave(&sdkp->zones_wp_offset_lock, flags);
379 for (zno = 0; zno < sdkp->zone_info.nr_zones; zno++) {
380 if (sdkp->zones_wp_offset[zno] != SD_ZBC_UPDATING_WP_OFST)
381 continue;
382
383 spin_unlock_irqrestore(&sdkp->zones_wp_offset_lock, flags);
384 ret = sd_zbc_do_report_zones(sdkp, sdkp->zone_wp_update_buf,
385 SD_BUF_SIZE,
386 zno * sdkp->zone_info.zone_blocks, true);
387 spin_lock_irqsave(&sdkp->zones_wp_offset_lock, flags);
388 if (!ret)
389 sd_zbc_parse_report(sdkp, sdkp->zone_wp_update_buf + 64,
390 zno, sd_zbc_update_wp_offset_cb,
391 sdkp);
392 }
393 spin_unlock_irqrestore(&sdkp->zones_wp_offset_lock, flags);
394
395 scsi_device_put(sdkp->device);
396}
397
398/**
399 * sd_zbc_prepare_zone_append() - Prepare an emulated ZONE_APPEND command.
400 * @cmd: the command to setup
401 * @lba: the LBA to patch
402 * @nr_blocks: the number of LBAs to be written
403 *
404 * Called from sd_setup_read_write_cmnd() for REQ_OP_ZONE_APPEND.
405 * @sd_zbc_prepare_zone_append() handles the necessary zone wrote locking and
406 * patching of the lba for an emulated ZONE_APPEND command.
407 *
408 * In case the cached write pointer offset is %SD_ZBC_INVALID_WP_OFST it will
409 * schedule a REPORT ZONES command and return BLK_STS_IOERR.
410 */
411blk_status_t sd_zbc_prepare_zone_append(struct scsi_cmnd *cmd, sector_t *lba,
412 unsigned int nr_blocks)
413{
414 struct request *rq = scsi_cmd_to_rq(cmd);
415 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
416 unsigned int wp_offset, zno = blk_rq_zone_no(rq);
417 unsigned long flags;
418 blk_status_t ret;
419
420 ret = sd_zbc_cmnd_checks(cmd);
421 if (ret != BLK_STS_OK)
422 return ret;
423
424 if (!blk_rq_zone_is_seq(rq))
425 return BLK_STS_IOERR;
426
427 /* Unlock of the write lock will happen in sd_zbc_complete() */
428 if (!blk_req_zone_write_trylock(rq))
429 return BLK_STS_ZONE_RESOURCE;
430
431 spin_lock_irqsave(&sdkp->zones_wp_offset_lock, flags);
432 wp_offset = sdkp->zones_wp_offset[zno];
433 switch (wp_offset) {
434 case SD_ZBC_INVALID_WP_OFST:
435 /*
436 * We are about to schedule work to update a zone write pointer
437 * offset, which will cause the zone append command to be
438 * requeued. So make sure that the scsi device does not go away
439 * while the work is being processed.
440 */
441 if (scsi_device_get(sdkp->device)) {
442 ret = BLK_STS_IOERR;
443 break;
444 }
445 sdkp->zones_wp_offset[zno] = SD_ZBC_UPDATING_WP_OFST;
446 schedule_work(&sdkp->zone_wp_offset_work);
447 fallthrough;
448 case SD_ZBC_UPDATING_WP_OFST:
449 ret = BLK_STS_DEV_RESOURCE;
450 break;
451 default:
452 wp_offset = sectors_to_logical(sdkp->device, wp_offset);
453 if (wp_offset + nr_blocks > sdkp->zone_info.zone_blocks) {
454 ret = BLK_STS_IOERR;
455 break;
456 }
457
458 trace_scsi_prepare_zone_append(cmd, *lba, wp_offset);
459 *lba += wp_offset;
460 }
461 spin_unlock_irqrestore(&sdkp->zones_wp_offset_lock, flags);
462 if (ret)
463 blk_req_zone_write_unlock(rq);
464 return ret;
465}
466
467/**
468 * sd_zbc_setup_zone_mgmt_cmnd - Prepare a zone ZBC_OUT command. The operations
469 * can be RESET WRITE POINTER, OPEN, CLOSE or FINISH.
470 * @cmd: the command to setup
471 * @op: Operation to be performed
472 * @all: All zones control
473 *
474 * Called from sd_init_command() for REQ_OP_ZONE_RESET, REQ_OP_ZONE_RESET_ALL,
475 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE or REQ_OP_ZONE_FINISH requests.
476 */
477blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
478 unsigned char op, bool all)
479{
480 struct request *rq = scsi_cmd_to_rq(cmd);
481 sector_t sector = blk_rq_pos(rq);
482 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
483 sector_t block = sectors_to_logical(sdkp->device, sector);
484 blk_status_t ret;
485
486 ret = sd_zbc_cmnd_checks(cmd);
487 if (ret != BLK_STS_OK)
488 return ret;
489
490 cmd->cmd_len = 16;
491 memset(cmd->cmnd, 0, cmd->cmd_len);
492 cmd->cmnd[0] = ZBC_OUT;
493 cmd->cmnd[1] = op;
494 if (all)
495 cmd->cmnd[14] = 0x1;
496 else
497 put_unaligned_be64(block, &cmd->cmnd[2]);
498
499 rq->timeout = SD_TIMEOUT;
500 cmd->sc_data_direction = DMA_NONE;
501 cmd->transfersize = 0;
502 cmd->allowed = 0;
503
504 return BLK_STS_OK;
505}
506
507static bool sd_zbc_need_zone_wp_update(struct request *rq)
508{
509 switch (req_op(rq)) {
510 case REQ_OP_ZONE_APPEND:
511 case REQ_OP_ZONE_FINISH:
512 case REQ_OP_ZONE_RESET:
513 case REQ_OP_ZONE_RESET_ALL:
514 return true;
515 case REQ_OP_WRITE:
516 case REQ_OP_WRITE_ZEROES:
517 return blk_rq_zone_is_seq(rq);
518 default:
519 return false;
520 }
521}
522
523/**
524 * sd_zbc_zone_wp_update - Update cached zone write pointer upon cmd completion
525 * @cmd: Completed command
526 * @good_bytes: Command reply bytes
527 *
528 * Called from sd_zbc_complete() to handle the update of the cached zone write
529 * pointer value in case an update is needed.
530 */
531static unsigned int sd_zbc_zone_wp_update(struct scsi_cmnd *cmd,
532 unsigned int good_bytes)
533{
534 int result = cmd->result;
535 struct request *rq = scsi_cmd_to_rq(cmd);
536 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
537 unsigned int zno = blk_rq_zone_no(rq);
538 enum req_op op = req_op(rq);
539 unsigned long flags;
540
541 /*
542 * If we got an error for a command that needs updating the write
543 * pointer offset cache, we must mark the zone wp offset entry as
544 * invalid to force an update from disk the next time a zone append
545 * command is issued.
546 */
547 spin_lock_irqsave(&sdkp->zones_wp_offset_lock, flags);
548
549 if (result && op != REQ_OP_ZONE_RESET_ALL) {
550 if (op == REQ_OP_ZONE_APPEND) {
551 /* Force complete completion (no retry) */
552 good_bytes = 0;
553 scsi_set_resid(cmd, blk_rq_bytes(rq));
554 }
555
556 /*
557 * Force an update of the zone write pointer offset on
558 * the next zone append access.
559 */
560 if (sdkp->zones_wp_offset[zno] != SD_ZBC_UPDATING_WP_OFST)
561 sdkp->zones_wp_offset[zno] = SD_ZBC_INVALID_WP_OFST;
562 goto unlock_wp_offset;
563 }
564
565 switch (op) {
566 case REQ_OP_ZONE_APPEND:
567 trace_scsi_zone_wp_update(cmd, rq->__sector,
568 sdkp->zones_wp_offset[zno], good_bytes);
569 rq->__sector += sdkp->zones_wp_offset[zno];
570 fallthrough;
571 case REQ_OP_WRITE_ZEROES:
572 case REQ_OP_WRITE:
573 if (sdkp->zones_wp_offset[zno] < sd_zbc_zone_sectors(sdkp))
574 sdkp->zones_wp_offset[zno] +=
575 good_bytes >> SECTOR_SHIFT;
576 break;
577 case REQ_OP_ZONE_RESET:
578 sdkp->zones_wp_offset[zno] = 0;
579 break;
580 case REQ_OP_ZONE_FINISH:
581 sdkp->zones_wp_offset[zno] = sd_zbc_zone_sectors(sdkp);
582 break;
583 case REQ_OP_ZONE_RESET_ALL:
584 memset(sdkp->zones_wp_offset, 0,
585 sdkp->zone_info.nr_zones * sizeof(unsigned int));
586 break;
587 default:
588 break;
589 }
590
591unlock_wp_offset:
592 spin_unlock_irqrestore(&sdkp->zones_wp_offset_lock, flags);
593
594 return good_bytes;
595}
596
597/**
598 * sd_zbc_complete - ZBC command post processing.
599 * @cmd: Completed command
600 * @good_bytes: Command reply bytes
601 * @sshdr: command sense header
602 *
603 * Called from sd_done() to handle zone commands errors and updates to the
604 * device queue zone write pointer offset cahce.
605 */
606unsigned int sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
607 struct scsi_sense_hdr *sshdr)
608{
609 int result = cmd->result;
610 struct request *rq = scsi_cmd_to_rq(cmd);
611
612 if (op_is_zone_mgmt(req_op(rq)) &&
613 result &&
614 sshdr->sense_key == ILLEGAL_REQUEST &&
615 sshdr->asc == 0x24) {
616 /*
617 * INVALID FIELD IN CDB error: a zone management command was
618 * attempted on a conventional zone. Nothing to worry about,
619 * so be quiet about the error.
620 */
621 rq->rq_flags |= RQF_QUIET;
622 } else if (sd_zbc_need_zone_wp_update(rq))
623 good_bytes = sd_zbc_zone_wp_update(cmd, good_bytes);
624
625 if (req_op(rq) == REQ_OP_ZONE_APPEND)
626 blk_req_zone_write_unlock(rq);
627
628 return good_bytes;
629}
630
631/**
632 * sd_zbc_check_zoned_characteristics - Check zoned block device characteristics
633 * @sdkp: Target disk
634 * @buf: Buffer where to store the VPD page data
635 *
636 * Read VPD page B6, get information and check that reads are unconstrained.
637 */
638static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
639 unsigned char *buf)
640{
641 u64 zone_starting_lba_gran;
642
643 if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
644 sd_printk(KERN_NOTICE, sdkp,
645 "Read zoned characteristics VPD page failed\n");
646 return -ENODEV;
647 }
648
649 if (sdkp->device->type != TYPE_ZBC) {
650 /* Host-aware */
651 sdkp->urswrz = 1;
652 sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]);
653 sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]);
654 sdkp->zones_max_open = 0;
655 return 0;
656 }
657
658 /* Host-managed */
659 sdkp->urswrz = buf[4] & 1;
660 sdkp->zones_optimal_open = 0;
661 sdkp->zones_optimal_nonseq = 0;
662 sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
663 /* Check zone alignment method */
664 switch (buf[23] & 0xf) {
665 case 0:
666 case ZBC_CONSTANT_ZONE_LENGTH:
667 /* Use zone length */
668 break;
669 case ZBC_CONSTANT_ZONE_START_OFFSET:
670 zone_starting_lba_gran = get_unaligned_be64(&buf[24]);
671 if (zone_starting_lba_gran == 0 ||
672 !is_power_of_2(zone_starting_lba_gran) ||
673 logical_to_sectors(sdkp->device, zone_starting_lba_gran) >
674 UINT_MAX) {
675 sd_printk(KERN_ERR, sdkp,
676 "Invalid zone starting LBA granularity %llu\n",
677 zone_starting_lba_gran);
678 return -ENODEV;
679 }
680 sdkp->zone_starting_lba_gran = zone_starting_lba_gran;
681 break;
682 default:
683 sd_printk(KERN_ERR, sdkp, "Invalid zone alignment method\n");
684 return -ENODEV;
685 }
686
687 /*
688 * Check for unconstrained reads: host-managed devices with
689 * constrained reads (drives failing read after write pointer)
690 * are not supported.
691 */
692 if (!sdkp->urswrz) {
693 if (sdkp->first_scan)
694 sd_printk(KERN_NOTICE, sdkp,
695 "constrained reads devices are not supported\n");
696 return -ENODEV;
697 }
698
699 return 0;
700}
701
702/**
703 * sd_zbc_check_capacity - Check the device capacity
704 * @sdkp: Target disk
705 * @buf: command buffer
706 * @zblocks: zone size in logical blocks
707 *
708 * Get the device zone size and check that the device capacity as reported
709 * by READ CAPACITY matches the max_lba value (plus one) of the report zones
710 * command reply for devices with RC_BASIS == 0.
711 *
712 * Returns 0 upon success or an error code upon failure.
713 */
714static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf,
715 u32 *zblocks)
716{
717 u64 zone_blocks;
718 sector_t max_lba;
719 unsigned char *rec;
720 int ret;
721
722 /* Do a report zone to get max_lba and the size of the first zone */
723 ret = sd_zbc_do_report_zones(sdkp, buf, SD_BUF_SIZE, 0, false);
724 if (ret)
725 return ret;
726
727 if (sdkp->rc_basis == 0) {
728 /* The max_lba field is the capacity of this device */
729 max_lba = get_unaligned_be64(&buf[8]);
730 if (sdkp->capacity != max_lba + 1) {
731 if (sdkp->first_scan)
732 sd_printk(KERN_WARNING, sdkp,
733 "Changing capacity from %llu to max LBA+1 %llu\n",
734 (unsigned long long)sdkp->capacity,
735 (unsigned long long)max_lba + 1);
736 sdkp->capacity = max_lba + 1;
737 }
738 }
739
740 if (sdkp->zone_starting_lba_gran == 0) {
741 /* Get the size of the first reported zone */
742 rec = buf + 64;
743 zone_blocks = get_unaligned_be64(&rec[8]);
744 if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
745 if (sdkp->first_scan)
746 sd_printk(KERN_NOTICE, sdkp,
747 "Zone size too large\n");
748 return -EFBIG;
749 }
750 } else {
751 zone_blocks = sdkp->zone_starting_lba_gran;
752 }
753
754 if (!is_power_of_2(zone_blocks)) {
755 sd_printk(KERN_ERR, sdkp,
756 "Zone size %llu is not a power of two.\n",
757 zone_blocks);
758 return -EINVAL;
759 }
760
761 *zblocks = zone_blocks;
762
763 return 0;
764}
765
766static void sd_zbc_print_zones(struct scsi_disk *sdkp)
767{
768 if (!sd_is_zoned(sdkp) || !sdkp->capacity)
769 return;
770
771 if (sdkp->capacity & (sdkp->zone_info.zone_blocks - 1))
772 sd_printk(KERN_NOTICE, sdkp,
773 "%u zones of %u logical blocks + 1 runt zone\n",
774 sdkp->zone_info.nr_zones - 1,
775 sdkp->zone_info.zone_blocks);
776 else
777 sd_printk(KERN_NOTICE, sdkp,
778 "%u zones of %u logical blocks\n",
779 sdkp->zone_info.nr_zones,
780 sdkp->zone_info.zone_blocks);
781}
782
783static int sd_zbc_init_disk(struct scsi_disk *sdkp)
784{
785 sdkp->zones_wp_offset = NULL;
786 spin_lock_init(&sdkp->zones_wp_offset_lock);
787 sdkp->rev_wp_offset = NULL;
788 mutex_init(&sdkp->rev_mutex);
789 INIT_WORK(&sdkp->zone_wp_offset_work, sd_zbc_update_wp_offset_workfn);
790 sdkp->zone_wp_update_buf = kzalloc(SD_BUF_SIZE, GFP_KERNEL);
791 if (!sdkp->zone_wp_update_buf)
792 return -ENOMEM;
793
794 return 0;
795}
796
797void sd_zbc_free_zone_info(struct scsi_disk *sdkp)
798{
799 if (!sdkp->zone_wp_update_buf)
800 return;
801
802 /* Serialize against revalidate zones */
803 mutex_lock(&sdkp->rev_mutex);
804
805 kvfree(sdkp->zones_wp_offset);
806 sdkp->zones_wp_offset = NULL;
807 kfree(sdkp->zone_wp_update_buf);
808 sdkp->zone_wp_update_buf = NULL;
809
810 sdkp->early_zone_info = (struct zoned_disk_info){ };
811 sdkp->zone_info = (struct zoned_disk_info){ };
812
813 mutex_unlock(&sdkp->rev_mutex);
814}
815
816static void sd_zbc_revalidate_zones_cb(struct gendisk *disk)
817{
818 struct scsi_disk *sdkp = scsi_disk(disk);
819
820 swap(sdkp->zones_wp_offset, sdkp->rev_wp_offset);
821}
822
823/*
824 * Call blk_revalidate_disk_zones() if any of the zoned disk properties have
825 * changed that make it necessary to call that function. Called by
826 * sd_revalidate_disk() after the gendisk capacity has been set.
827 */
828int sd_zbc_revalidate_zones(struct scsi_disk *sdkp)
829{
830 struct gendisk *disk = sdkp->disk;
831 struct request_queue *q = disk->queue;
832 u32 zone_blocks = sdkp->early_zone_info.zone_blocks;
833 unsigned int nr_zones = sdkp->early_zone_info.nr_zones;
834 int ret = 0;
835 unsigned int flags;
836
837 /*
838 * For all zoned disks, initialize zone append emulation data if not
839 * already done.
840 */
841 if (sd_is_zoned(sdkp) && !sdkp->zone_wp_update_buf) {
842 ret = sd_zbc_init_disk(sdkp);
843 if (ret)
844 return ret;
845 }
846
847 /*
848 * There is nothing to do for regular disks, including host-aware disks
849 * that have partitions.
850 */
851 if (!blk_queue_is_zoned(q))
852 return 0;
853
854 /*
855 * Make sure revalidate zones are serialized to ensure exclusive
856 * updates of the scsi disk data.
857 */
858 mutex_lock(&sdkp->rev_mutex);
859
860 if (sdkp->zone_info.zone_blocks == zone_blocks &&
861 sdkp->zone_info.nr_zones == nr_zones &&
862 disk->nr_zones == nr_zones)
863 goto unlock;
864
865 flags = memalloc_noio_save();
866 sdkp->zone_info.zone_blocks = zone_blocks;
867 sdkp->zone_info.nr_zones = nr_zones;
868 sdkp->rev_wp_offset = kvcalloc(nr_zones, sizeof(u32), GFP_KERNEL);
869 if (!sdkp->rev_wp_offset) {
870 ret = -ENOMEM;
871 memalloc_noio_restore(flags);
872 goto unlock;
873 }
874
875 blk_queue_chunk_sectors(q,
876 logical_to_sectors(sdkp->device, zone_blocks));
877 blk_queue_max_zone_append_sectors(q,
878 q->limits.max_segments << PAGE_SECTORS_SHIFT);
879
880 ret = blk_revalidate_disk_zones(disk, sd_zbc_revalidate_zones_cb);
881
882 memalloc_noio_restore(flags);
883 kvfree(sdkp->rev_wp_offset);
884 sdkp->rev_wp_offset = NULL;
885
886 if (ret) {
887 sdkp->zone_info = (struct zoned_disk_info){ };
888 sdkp->capacity = 0;
889 goto unlock;
890 }
891
892 sd_zbc_print_zones(sdkp);
893
894unlock:
895 mutex_unlock(&sdkp->rev_mutex);
896
897 return ret;
898}
899
900/**
901 * sd_zbc_read_zones - Read zone information and update the request queue
902 * @sdkp: SCSI disk pointer.
903 * @buf: 512 byte buffer used for storing SCSI command output.
904 *
905 * Read zone information and update the request queue zone characteristics and
906 * also the zoned device information in *sdkp. Called by sd_revalidate_disk()
907 * before the gendisk capacity has been set.
908 */
909int sd_zbc_read_zones(struct scsi_disk *sdkp, u8 buf[SD_BUF_SIZE])
910{
911 struct gendisk *disk = sdkp->disk;
912 struct request_queue *q = disk->queue;
913 unsigned int nr_zones;
914 u32 zone_blocks = 0;
915 int ret;
916
917 if (!sd_is_zoned(sdkp)) {
918 /*
919 * Device managed or normal SCSI disk, no special handling
920 * required. Nevertheless, free the disk zone information in
921 * case the device type changed.
922 */
923 sd_zbc_free_zone_info(sdkp);
924 return 0;
925 }
926
927 /* READ16/WRITE16/SYNC16 is mandatory for ZBC devices */
928 sdkp->device->use_16_for_rw = 1;
929 sdkp->device->use_10_for_rw = 0;
930 sdkp->device->use_16_for_sync = 1;
931
932 /* Check zoned block device characteristics (unconstrained reads) */
933 ret = sd_zbc_check_zoned_characteristics(sdkp, buf);
934 if (ret)
935 goto err;
936
937 /* Check the device capacity reported by report zones */
938 ret = sd_zbc_check_capacity(sdkp, buf, &zone_blocks);
939 if (ret != 0)
940 goto err;
941
942 /* The drive satisfies the kernel restrictions: set it up */
943 blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
944 blk_queue_required_elevator_features(q, ELEVATOR_F_ZBD_SEQ_WRITE);
945 if (sdkp->zones_max_open == U32_MAX)
946 disk_set_max_open_zones(disk, 0);
947 else
948 disk_set_max_open_zones(disk, sdkp->zones_max_open);
949 disk_set_max_active_zones(disk, 0);
950 nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
951
952 sdkp->early_zone_info.nr_zones = nr_zones;
953 sdkp->early_zone_info.zone_blocks = zone_blocks;
954
955 return 0;
956
957err:
958 sdkp->capacity = 0;
959
960 return ret;
961}
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
32#include "sd.h"
33
34/**
35 * sd_zbc_parse_report - Convert a zone descriptor to a struct blk_zone,
36 * @sdkp: The disk the report originated from
37 * @buf: Address of the report zone descriptor
38 * @zone: the destination zone structure
39 *
40 * All LBA sized values are converted to 512B sectors unit.
41 */
42static void sd_zbc_parse_report(struct scsi_disk *sdkp, u8 *buf,
43 struct blk_zone *zone)
44{
45 struct scsi_device *sdp = sdkp->device;
46
47 memset(zone, 0, sizeof(struct blk_zone));
48
49 zone->type = buf[0] & 0x0f;
50 zone->cond = (buf[1] >> 4) & 0xf;
51 if (buf[1] & 0x01)
52 zone->reset = 1;
53 if (buf[1] & 0x02)
54 zone->non_seq = 1;
55
56 zone->len = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
57 zone->start = logical_to_sectors(sdp, get_unaligned_be64(&buf[16]));
58 zone->wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24]));
59 if (zone->type != ZBC_ZONE_TYPE_CONV &&
60 zone->cond == ZBC_ZONE_COND_FULL)
61 zone->wp = zone->start + zone->len;
62}
63
64/**
65 * sd_zbc_report_zones - Issue a REPORT ZONES scsi command.
66 * @sdkp: The target disk
67 * @buf: Buffer to use for the reply
68 * @buflen: the buffer size
69 * @lba: Start LBA of the report
70 *
71 * For internal use during device validation.
72 */
73static int sd_zbc_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
74 unsigned int buflen, sector_t lba)
75{
76 struct scsi_device *sdp = sdkp->device;
77 const int timeout = sdp->request_queue->rq_timeout;
78 struct scsi_sense_hdr sshdr;
79 unsigned char cmd[16];
80 unsigned int rep_len;
81 int result;
82
83 memset(cmd, 0, 16);
84 cmd[0] = ZBC_IN;
85 cmd[1] = ZI_REPORT_ZONES;
86 put_unaligned_be64(lba, &cmd[2]);
87 put_unaligned_be32(buflen, &cmd[10]);
88 memset(buf, 0, buflen);
89
90 result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
91 buf, buflen, &sshdr,
92 timeout, SD_MAX_RETRIES, NULL);
93 if (result) {
94 sd_printk(KERN_ERR, sdkp,
95 "REPORT ZONES lba %llu failed with %d/%d\n",
96 (unsigned long long)lba,
97 host_byte(result), driver_byte(result));
98 return -EIO;
99 }
100
101 rep_len = get_unaligned_be32(&buf[0]);
102 if (rep_len < 64) {
103 sd_printk(KERN_ERR, sdkp,
104 "REPORT ZONES report invalid length %u\n",
105 rep_len);
106 return -EIO;
107 }
108
109 return 0;
110}
111
112/**
113 * sd_zbc_setup_report_cmnd - Prepare a REPORT ZONES scsi command
114 * @cmd: The command to setup
115 *
116 * Call in sd_init_command() for a REQ_OP_ZONE_REPORT request.
117 */
118int sd_zbc_setup_report_cmnd(struct scsi_cmnd *cmd)
119{
120 struct request *rq = cmd->request;
121 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
122 sector_t lba, sector = blk_rq_pos(rq);
123 unsigned int nr_bytes = blk_rq_bytes(rq);
124 int ret;
125
126 WARN_ON(nr_bytes == 0);
127
128 if (!sd_is_zoned(sdkp))
129 /* Not a zoned device */
130 return BLKPREP_KILL;
131
132 ret = scsi_init_io(cmd);
133 if (ret != BLKPREP_OK)
134 return ret;
135
136 cmd->cmd_len = 16;
137 memset(cmd->cmnd, 0, cmd->cmd_len);
138 cmd->cmnd[0] = ZBC_IN;
139 cmd->cmnd[1] = ZI_REPORT_ZONES;
140 lba = sectors_to_logical(sdkp->device, sector);
141 put_unaligned_be64(lba, &cmd->cmnd[2]);
142 put_unaligned_be32(nr_bytes, &cmd->cmnd[10]);
143 /* Do partial report for speeding things up */
144 cmd->cmnd[14] = ZBC_REPORT_ZONE_PARTIAL;
145
146 cmd->sc_data_direction = DMA_FROM_DEVICE;
147 cmd->sdb.length = nr_bytes;
148 cmd->transfersize = sdkp->device->sector_size;
149 cmd->allowed = 0;
150
151 /*
152 * Report may return less bytes than requested. Make sure
153 * to report completion on the entire initial request.
154 */
155 rq->__data_len = nr_bytes;
156
157 return BLKPREP_OK;
158}
159
160/**
161 * sd_zbc_report_zones_complete - Process a REPORT ZONES scsi command reply.
162 * @scmd: The completed report zones command
163 * @good_bytes: reply size in bytes
164 *
165 * Convert all reported zone descriptors to struct blk_zone. The conversion
166 * is done in-place, directly in the request specified sg buffer.
167 */
168static 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/**
224 * sd_zbc_zone_sectors - Get the device zone size in number of 512B sectors.
225 * @sdkp: The target disk
226 */
227static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
228{
229 return logical_to_sectors(sdkp->device, sdkp->zone_blocks);
230}
231
232/**
233 * sd_zbc_setup_reset_cmnd - Prepare a RESET WRITE POINTER scsi command.
234 * @cmd: the command to setup
235 *
236 * Called from sd_init_command() for a REQ_OP_ZONE_RESET request.
237 */
238int sd_zbc_setup_reset_cmnd(struct scsi_cmnd *cmd)
239{
240 struct request *rq = cmd->request;
241 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
242 sector_t sector = blk_rq_pos(rq);
243 sector_t block = sectors_to_logical(sdkp->device, sector);
244
245 if (!sd_is_zoned(sdkp))
246 /* Not a zoned device */
247 return BLKPREP_KILL;
248
249 if (sdkp->device->changed)
250 return BLKPREP_KILL;
251
252 if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
253 /* Unaligned request */
254 return BLKPREP_KILL;
255
256 cmd->cmd_len = 16;
257 memset(cmd->cmnd, 0, cmd->cmd_len);
258 cmd->cmnd[0] = ZBC_OUT;
259 cmd->cmnd[1] = ZO_RESET_WRITE_POINTER;
260 put_unaligned_be64(block, &cmd->cmnd[2]);
261
262 rq->timeout = SD_TIMEOUT;
263 cmd->sc_data_direction = DMA_NONE;
264 cmd->transfersize = 0;
265 cmd->allowed = 0;
266
267 return BLKPREP_OK;
268}
269
270/**
271 * sd_zbc_complete - ZBC command post processing.
272 * @cmd: Completed command
273 * @good_bytes: Command reply bytes
274 * @sshdr: command sense header
275 *
276 * Called from sd_done(). Process report zones reply and handle reset zone
277 * and write commands errors.
278 */
279void sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
280 struct scsi_sense_hdr *sshdr)
281{
282 int result = cmd->result;
283 struct request *rq = cmd->request;
284
285 switch (req_op(rq)) {
286 case REQ_OP_ZONE_RESET:
287
288 if (result &&
289 sshdr->sense_key == ILLEGAL_REQUEST &&
290 sshdr->asc == 0x24)
291 /*
292 * INVALID FIELD IN CDB error: reset of a conventional
293 * zone was attempted. Nothing to worry about, so be
294 * quiet about the error.
295 */
296 rq->rq_flags |= RQF_QUIET;
297 break;
298
299 case REQ_OP_WRITE:
300 case REQ_OP_WRITE_ZEROES:
301 case REQ_OP_WRITE_SAME:
302
303 if (result &&
304 sshdr->sense_key == ILLEGAL_REQUEST &&
305 sshdr->asc == 0x21)
306 /*
307 * INVALID ADDRESS FOR WRITE error: It is unlikely that
308 * retrying write requests failed with any kind of
309 * alignement error will result in success. So don't.
310 */
311 cmd->allowed = 0;
312 break;
313
314 case REQ_OP_ZONE_REPORT:
315
316 if (!result)
317 sd_zbc_report_zones_complete(cmd, good_bytes);
318 break;
319
320 }
321}
322
323/**
324 * sd_zbc_read_zoned_characteristics - Read zoned block device characteristics
325 * @sdkp: Target disk
326 * @buf: Buffer where to store the VPD page data
327 *
328 * Read VPD page B6.
329 */
330static int sd_zbc_read_zoned_characteristics(struct scsi_disk *sdkp,
331 unsigned char *buf)
332{
333
334 if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
335 sd_printk(KERN_NOTICE, sdkp,
336 "Unconstrained-read check failed\n");
337 return -ENODEV;
338 }
339
340 if (sdkp->device->type != TYPE_ZBC) {
341 /* Host-aware */
342 sdkp->urswrz = 1;
343 sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]);
344 sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]);
345 sdkp->zones_max_open = 0;
346 } else {
347 /* Host-managed */
348 sdkp->urswrz = buf[4] & 1;
349 sdkp->zones_optimal_open = 0;
350 sdkp->zones_optimal_nonseq = 0;
351 sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
352 }
353
354 return 0;
355}
356
357/**
358 * sd_zbc_check_capacity - Check reported capacity.
359 * @sdkp: Target disk
360 * @buf: Buffer to use for commands
361 *
362 * ZBC drive may report only the capacity of the first conventional zones at
363 * LBA 0. This is indicated by the RC_BASIS field of the read capacity reply.
364 * Check this here. If the disk reported only its conventional zones capacity,
365 * get the total capacity by doing a report zones.
366 */
367static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf)
368{
369 sector_t lba;
370 int ret;
371
372 if (sdkp->rc_basis != 0)
373 return 0;
374
375 /* Do a report zone to get the maximum LBA to check capacity */
376 ret = sd_zbc_report_zones(sdkp, buf, SD_BUF_SIZE, 0);
377 if (ret)
378 return ret;
379
380 /* The max_lba field is the capacity of this device */
381 lba = get_unaligned_be64(&buf[8]);
382 if (lba + 1 == sdkp->capacity)
383 return 0;
384
385 if (sdkp->first_scan)
386 sd_printk(KERN_WARNING, sdkp,
387 "Changing capacity from %llu to max LBA+1 %llu\n",
388 (unsigned long long)sdkp->capacity,
389 (unsigned long long)lba + 1);
390 sdkp->capacity = lba + 1;
391
392 return 0;
393}
394
395#define SD_ZBC_BUF_SIZE 131072U
396
397/**
398 * sd_zbc_check_zone_size - Check the device zone sizes
399 * @sdkp: Target disk
400 *
401 * Check that all zones of the device are equal. The last zone can however
402 * be smaller. The zone size must also be a power of two number of LBAs.
403 *
404 * Returns the zone size in bytes upon success or an error code upon failure.
405 */
406static s64 sd_zbc_check_zone_size(struct scsi_disk *sdkp)
407{
408 u64 zone_blocks = 0;
409 sector_t block = 0;
410 unsigned char *buf;
411 unsigned char *rec;
412 unsigned int buf_len;
413 unsigned int list_length;
414 int ret;
415 u8 same;
416
417 /* Get a buffer */
418 buf = kmalloc(SD_ZBC_BUF_SIZE, GFP_KERNEL);
419 if (!buf)
420 return -ENOMEM;
421
422 /* Do a report zone to get the same field */
423 ret = sd_zbc_report_zones(sdkp, buf, SD_ZBC_BUF_SIZE, 0);
424 if (ret)
425 goto out_free;
426
427 same = buf[4] & 0x0f;
428 if (same > 0) {
429 rec = &buf[64];
430 zone_blocks = get_unaligned_be64(&rec[8]);
431 goto out;
432 }
433
434 /*
435 * Check the size of all zones: all zones must be of
436 * equal size, except the last zone which can be smaller
437 * than other zones.
438 */
439 do {
440
441 /* Parse REPORT ZONES header */
442 list_length = get_unaligned_be32(&buf[0]) + 64;
443 rec = buf + 64;
444 buf_len = min(list_length, SD_ZBC_BUF_SIZE);
445
446 /* Parse zone descriptors */
447 while (rec < buf + buf_len) {
448 u64 this_zone_blocks = get_unaligned_be64(&rec[8]);
449
450 if (zone_blocks == 0) {
451 zone_blocks = this_zone_blocks;
452 } else if (this_zone_blocks != zone_blocks &&
453 (block + this_zone_blocks < sdkp->capacity
454 || this_zone_blocks > zone_blocks)) {
455 this_zone_blocks = 0;
456 goto out;
457 }
458 block += this_zone_blocks;
459 rec += 64;
460 }
461
462 if (block < sdkp->capacity) {
463 ret = sd_zbc_report_zones(sdkp, buf,
464 SD_ZBC_BUF_SIZE, block);
465 if (ret)
466 goto out_free;
467 }
468
469 } while (block < sdkp->capacity);
470
471out:
472 if (!zone_blocks) {
473 if (sdkp->first_scan)
474 sd_printk(KERN_NOTICE, sdkp,
475 "Devices with non constant zone "
476 "size are not supported\n");
477 ret = -ENODEV;
478 } else if (!is_power_of_2(zone_blocks)) {
479 if (sdkp->first_scan)
480 sd_printk(KERN_NOTICE, sdkp,
481 "Devices with non power of 2 zone "
482 "size are not supported\n");
483 ret = -ENODEV;
484 } else if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
485 if (sdkp->first_scan)
486 sd_printk(KERN_NOTICE, sdkp,
487 "Zone size too large\n");
488 ret = -ENODEV;
489 } else {
490 ret = zone_blocks;
491 }
492
493out_free:
494 kfree(buf);
495
496 return ret;
497}
498
499/**
500 * sd_zbc_alloc_zone_bitmap - Allocate a zone bitmap (one bit per zone).
501 * @nr_zones: Number of zones to allocate space for.
502 * @numa_node: NUMA node to allocate the memory from.
503 */
504static inline unsigned long *
505sd_zbc_alloc_zone_bitmap(u32 nr_zones, int numa_node)
506{
507 return kzalloc_node(BITS_TO_LONGS(nr_zones) * sizeof(unsigned long),
508 GFP_KERNEL, numa_node);
509}
510
511/**
512 * sd_zbc_get_seq_zones - Parse report zones reply to identify sequential zones
513 * @sdkp: disk used
514 * @buf: report reply buffer
515 * @buflen: length of @buf
516 * @zone_shift: logarithm base 2 of the number of blocks in a zone
517 * @seq_zones_bitmap: bitmap of sequential zones to set
518 *
519 * Parse reported zone descriptors in @buf to identify sequential zones and
520 * set the reported zone bit in @seq_zones_bitmap accordingly.
521 * Since read-only and offline zones cannot be written, do not
522 * mark them as sequential in the bitmap.
523 * Return the LBA after the last zone reported.
524 */
525static sector_t sd_zbc_get_seq_zones(struct scsi_disk *sdkp, unsigned char *buf,
526 unsigned int buflen, u32 zone_shift,
527 unsigned long *seq_zones_bitmap)
528{
529 sector_t lba, next_lba = sdkp->capacity;
530 unsigned int buf_len, list_length;
531 unsigned char *rec;
532 u8 type, cond;
533
534 list_length = get_unaligned_be32(&buf[0]) + 64;
535 buf_len = min(list_length, buflen);
536 rec = buf + 64;
537
538 while (rec < buf + buf_len) {
539 type = rec[0] & 0x0f;
540 cond = (rec[1] >> 4) & 0xf;
541 lba = get_unaligned_be64(&rec[16]);
542 if (type != ZBC_ZONE_TYPE_CONV &&
543 cond != ZBC_ZONE_COND_READONLY &&
544 cond != ZBC_ZONE_COND_OFFLINE)
545 set_bit(lba >> zone_shift, seq_zones_bitmap);
546 next_lba = lba + get_unaligned_be64(&rec[8]);
547 rec += 64;
548 }
549
550 return next_lba;
551}
552
553/**
554 * sd_zbc_setup_seq_zones_bitmap - Initialize a seq zone bitmap.
555 * @sdkp: target disk
556 * @zone_shift: logarithm base 2 of the number of blocks in a zone
557 * @nr_zones: number of zones to set up a seq zone bitmap for
558 *
559 * Allocate a zone bitmap and initialize it by identifying sequential zones.
560 */
561static unsigned long *
562sd_zbc_setup_seq_zones_bitmap(struct scsi_disk *sdkp, u32 zone_shift,
563 u32 nr_zones)
564{
565 struct request_queue *q = sdkp->disk->queue;
566 unsigned long *seq_zones_bitmap;
567 sector_t lba = 0;
568 unsigned char *buf;
569 int ret = -ENOMEM;
570
571 seq_zones_bitmap = sd_zbc_alloc_zone_bitmap(nr_zones, q->node);
572 if (!seq_zones_bitmap)
573 return ERR_PTR(-ENOMEM);
574
575 buf = kmalloc(SD_ZBC_BUF_SIZE, GFP_KERNEL);
576 if (!buf)
577 goto out;
578
579 while (lba < sdkp->capacity) {
580 ret = sd_zbc_report_zones(sdkp, buf, SD_ZBC_BUF_SIZE, lba);
581 if (ret)
582 goto out;
583 lba = sd_zbc_get_seq_zones(sdkp, buf, SD_ZBC_BUF_SIZE,
584 zone_shift, seq_zones_bitmap);
585 }
586
587 if (lba != sdkp->capacity) {
588 /* Something went wrong */
589 ret = -EIO;
590 }
591
592out:
593 kfree(buf);
594 if (ret) {
595 kfree(seq_zones_bitmap);
596 return ERR_PTR(ret);
597 }
598 return seq_zones_bitmap;
599}
600
601static void sd_zbc_cleanup(struct scsi_disk *sdkp)
602{
603 struct request_queue *q = sdkp->disk->queue;
604
605 kfree(q->seq_zones_bitmap);
606 q->seq_zones_bitmap = NULL;
607
608 kfree(q->seq_zones_wlock);
609 q->seq_zones_wlock = NULL;
610
611 q->nr_zones = 0;
612}
613
614static int sd_zbc_setup(struct scsi_disk *sdkp, u32 zone_blocks)
615{
616 struct request_queue *q = sdkp->disk->queue;
617 u32 zone_shift = ilog2(zone_blocks);
618 u32 nr_zones;
619 int ret;
620
621 /* chunk_sectors indicates the zone size */
622 blk_queue_chunk_sectors(q,
623 logical_to_sectors(sdkp->device, zone_blocks));
624 nr_zones = round_up(sdkp->capacity, zone_blocks) >> zone_shift;
625
626 /*
627 * Initialize the device request queue information if the number
628 * of zones changed.
629 */
630 if (nr_zones != sdkp->nr_zones || nr_zones != q->nr_zones) {
631 unsigned long *seq_zones_wlock = NULL, *seq_zones_bitmap = NULL;
632 size_t zone_bitmap_size;
633
634 if (nr_zones) {
635 seq_zones_wlock = sd_zbc_alloc_zone_bitmap(nr_zones,
636 q->node);
637 if (!seq_zones_wlock) {
638 ret = -ENOMEM;
639 goto err;
640 }
641
642 seq_zones_bitmap = sd_zbc_setup_seq_zones_bitmap(sdkp,
643 zone_shift, nr_zones);
644 if (IS_ERR(seq_zones_bitmap)) {
645 ret = PTR_ERR(seq_zones_bitmap);
646 kfree(seq_zones_wlock);
647 goto err;
648 }
649 }
650 zone_bitmap_size = BITS_TO_LONGS(nr_zones) *
651 sizeof(unsigned long);
652 blk_mq_freeze_queue(q);
653 if (q->nr_zones != nr_zones) {
654 /* READ16/WRITE16 is mandatory for ZBC disks */
655 sdkp->device->use_16_for_rw = 1;
656 sdkp->device->use_10_for_rw = 0;
657
658 sdkp->zone_blocks = zone_blocks;
659 sdkp->zone_shift = zone_shift;
660 sdkp->nr_zones = nr_zones;
661 q->nr_zones = nr_zones;
662 swap(q->seq_zones_wlock, seq_zones_wlock);
663 swap(q->seq_zones_bitmap, seq_zones_bitmap);
664 } else if (memcmp(q->seq_zones_bitmap, seq_zones_bitmap,
665 zone_bitmap_size) != 0) {
666 memcpy(q->seq_zones_bitmap, seq_zones_bitmap,
667 zone_bitmap_size);
668 }
669 blk_mq_unfreeze_queue(q);
670 kfree(seq_zones_wlock);
671 kfree(seq_zones_bitmap);
672 }
673
674 return 0;
675
676err:
677 sd_zbc_cleanup(sdkp);
678 return ret;
679}
680
681int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
682{
683 int64_t zone_blocks;
684 int ret;
685
686 if (!sd_is_zoned(sdkp))
687 /*
688 * Device managed or normal SCSI disk,
689 * no special handling required
690 */
691 return 0;
692
693 /* Get zoned block device characteristics */
694 ret = sd_zbc_read_zoned_characteristics(sdkp, buf);
695 if (ret)
696 goto err;
697
698 /*
699 * Check for unconstrained reads: host-managed devices with
700 * constrained reads (drives failing read after write pointer)
701 * are not supported.
702 */
703 if (!sdkp->urswrz) {
704 if (sdkp->first_scan)
705 sd_printk(KERN_NOTICE, sdkp,
706 "constrained reads devices are not supported\n");
707 ret = -ENODEV;
708 goto err;
709 }
710
711 /* Check capacity */
712 ret = sd_zbc_check_capacity(sdkp, buf);
713 if (ret)
714 goto err;
715
716 /*
717 * Check zone size: only devices with a constant zone size (except
718 * an eventual last runt zone) that is a power of 2 are supported.
719 */
720 zone_blocks = sd_zbc_check_zone_size(sdkp);
721 ret = -EFBIG;
722 if (zone_blocks != (u32)zone_blocks)
723 goto err;
724 ret = zone_blocks;
725 if (ret < 0)
726 goto err;
727
728 /* The drive satisfies the kernel restrictions: set it up */
729 ret = sd_zbc_setup(sdkp, zone_blocks);
730 if (ret)
731 goto err;
732
733 return 0;
734
735err:
736 sdkp->capacity = 0;
737 sd_zbc_cleanup(sdkp);
738
739 return ret;
740}
741
742void sd_zbc_remove(struct scsi_disk *sdkp)
743{
744 sd_zbc_cleanup(sdkp);
745}
746
747void sd_zbc_print_zones(struct scsi_disk *sdkp)
748{
749 if (!sd_is_zoned(sdkp) || !sdkp->capacity)
750 return;
751
752 if (sdkp->capacity & (sdkp->zone_blocks - 1))
753 sd_printk(KERN_NOTICE, sdkp,
754 "%u zones of %u logical blocks + 1 runt zone\n",
755 sdkp->nr_zones - 1,
756 sdkp->zone_blocks);
757 else
758 sd_printk(KERN_NOTICE, sdkp,
759 "%u zones of %u logical blocks\n",
760 sdkp->nr_zones,
761 sdkp->zone_blocks);
762}