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