]> git.ipfire.org Git - thirdparty/linux.git/blob - drivers/scsi/sd_zbc.c
Merge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
[thirdparty/linux.git] / drivers / scsi / sd_zbc.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * SCSI Zoned Block commands
4 *
5 * Copyright (C) 2014-2015 SUSE Linux GmbH
6 * Written by: Hannes Reinecke <hare@suse.de>
7 * Modified by: Damien Le Moal <damien.lemoal@hgst.com>
8 * Modified by: Shaun Tancheff <shaun.tancheff@seagate.com>
9 */
10
11 #include <linux/blkdev.h>
12 #include <linux/vmalloc.h>
13 #include <linux/sched/mm.h>
14
15 #include <asm/unaligned.h>
16
17 #include <scsi/scsi.h>
18 #include <scsi/scsi_cmnd.h>
19
20 #include "sd.h"
21
22 /**
23 * sd_zbc_parse_report - Convert a zone descriptor to a struct blk_zone,
24 * @sdkp: The disk the report originated from
25 * @buf: Address of the report zone descriptor
26 * @zone: the destination zone structure
27 *
28 * All LBA sized values are converted to 512B sectors unit.
29 */
30 static void sd_zbc_parse_report(struct scsi_disk *sdkp, u8 *buf,
31 struct blk_zone *zone)
32 {
33 struct scsi_device *sdp = sdkp->device;
34
35 memset(zone, 0, sizeof(struct blk_zone));
36
37 zone->type = buf[0] & 0x0f;
38 zone->cond = (buf[1] >> 4) & 0xf;
39 if (buf[1] & 0x01)
40 zone->reset = 1;
41 if (buf[1] & 0x02)
42 zone->non_seq = 1;
43
44 zone->len = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
45 zone->start = logical_to_sectors(sdp, get_unaligned_be64(&buf[16]));
46 zone->wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24]));
47 if (zone->type != ZBC_ZONE_TYPE_CONV &&
48 zone->cond == ZBC_ZONE_COND_FULL)
49 zone->wp = zone->start + zone->len;
50 }
51
52 /**
53 * sd_zbc_do_report_zones - Issue a REPORT ZONES scsi command.
54 * @sdkp: The target disk
55 * @buf: vmalloc-ed buffer to use for the reply
56 * @buflen: the buffer size
57 * @lba: Start LBA of the report
58 * @partial: Do partial report
59 *
60 * For internal use during device validation.
61 * Using partial=true can significantly speed up execution of a report zones
62 * command because the disk does not have to count all possible report matching
63 * zones and will only report the count of zones fitting in the command reply
64 * buffer.
65 */
66 static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
67 unsigned int buflen, sector_t lba,
68 bool partial)
69 {
70 struct scsi_device *sdp = sdkp->device;
71 const int timeout = sdp->request_queue->rq_timeout;
72 struct scsi_sense_hdr sshdr;
73 unsigned char cmd[16];
74 unsigned int rep_len;
75 int result;
76
77 memset(cmd, 0, 16);
78 cmd[0] = ZBC_IN;
79 cmd[1] = ZI_REPORT_ZONES;
80 put_unaligned_be64(lba, &cmd[2]);
81 put_unaligned_be32(buflen, &cmd[10]);
82 if (partial)
83 cmd[14] = ZBC_REPORT_ZONE_PARTIAL;
84
85 result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
86 buf, buflen, &sshdr,
87 timeout, SD_MAX_RETRIES, NULL);
88 if (result) {
89 sd_printk(KERN_ERR, sdkp,
90 "REPORT ZONES lba %llu failed with %d/%d\n",
91 (unsigned long long)lba,
92 host_byte(result), driver_byte(result));
93 return -EIO;
94 }
95
96 rep_len = get_unaligned_be32(&buf[0]);
97 if (rep_len < 64) {
98 sd_printk(KERN_ERR, sdkp,
99 "REPORT ZONES report invalid length %u\n",
100 rep_len);
101 return -EIO;
102 }
103
104 return 0;
105 }
106
107 /*
108 * Maximum number of zones to get with one report zones command.
109 */
110 #define SD_ZBC_REPORT_MAX_ZONES 8192U
111
112 /**
113 * Allocate a buffer for report zones reply.
114 * @sdkp: The target disk
115 * @nr_zones: Maximum number of zones to report
116 * @buflen: Size of the buffer allocated
117 *
118 * Try to allocate a reply buffer for the number of requested zones.
119 * The size of the buffer allocated may be smaller than requested to
120 * satify the device constraint (max_hw_sectors, max_segments, etc).
121 *
122 * Return the address of the allocated buffer and update @buflen with
123 * the size of the allocated buffer.
124 */
125 static void *sd_zbc_alloc_report_buffer(struct scsi_disk *sdkp,
126 unsigned int nr_zones, size_t *buflen)
127 {
128 struct request_queue *q = sdkp->disk->queue;
129 size_t bufsize;
130 void *buf;
131
132 /*
133 * Report zone buffer size should be at most 64B times the number of
134 * zones requested plus the 64B reply header, but should be at least
135 * SECTOR_SIZE for ATA devices.
136 * Make sure that this size does not exceed the hardware capabilities.
137 * Furthermore, since the report zone command cannot be split, make
138 * sure that the allocated buffer can always be mapped by limiting the
139 * number of pages allocated to the HBA max segments limit.
140 */
141 nr_zones = min(nr_zones, SD_ZBC_REPORT_MAX_ZONES);
142 bufsize = roundup((nr_zones + 1) * 64, 512);
143 bufsize = min_t(size_t, bufsize,
144 queue_max_hw_sectors(q) << SECTOR_SHIFT);
145 bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
146
147 buf = vzalloc(bufsize);
148 if (buf)
149 *buflen = bufsize;
150
151 return buf;
152 }
153
154 /**
155 * sd_zbc_report_zones - Disk report zones operation.
156 * @disk: The target disk
157 * @sector: Start 512B sector of the report
158 * @zones: Array of zone descriptors
159 * @nr_zones: Number of descriptors in the array
160 *
161 * Execute a report zones command on the target disk.
162 */
163 int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
164 struct blk_zone *zones, unsigned int *nr_zones)
165 {
166 struct scsi_disk *sdkp = scsi_disk(disk);
167 unsigned int i, nrz = *nr_zones;
168 unsigned char *buf;
169 size_t buflen = 0, offset = 0;
170 int ret = 0;
171
172 if (!sd_is_zoned(sdkp))
173 /* Not a zoned device */
174 return -EOPNOTSUPP;
175
176 buf = sd_zbc_alloc_report_buffer(sdkp, nrz, &buflen);
177 if (!buf)
178 return -ENOMEM;
179
180 ret = sd_zbc_do_report_zones(sdkp, buf, buflen,
181 sectors_to_logical(sdkp->device, sector), true);
182 if (ret)
183 goto out;
184
185 nrz = min(nrz, get_unaligned_be32(&buf[0]) / 64);
186 for (i = 0; i < nrz; i++) {
187 offset += 64;
188 sd_zbc_parse_report(sdkp, buf + offset, zones);
189 zones++;
190 }
191
192 *nr_zones = nrz;
193
194 out:
195 kvfree(buf);
196
197 return ret;
198 }
199
200 /**
201 * sd_zbc_zone_sectors - Get the device zone size in number of 512B sectors.
202 * @sdkp: The target disk
203 */
204 static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
205 {
206 return logical_to_sectors(sdkp->device, sdkp->zone_blocks);
207 }
208
209 /**
210 * sd_zbc_setup_reset_cmnd - Prepare a RESET WRITE POINTER scsi command.
211 * @cmd: the command to setup
212 *
213 * Called from sd_init_command() for a REQ_OP_ZONE_RESET request.
214 */
215 blk_status_t sd_zbc_setup_reset_cmnd(struct scsi_cmnd *cmd)
216 {
217 struct request *rq = cmd->request;
218 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
219 sector_t sector = blk_rq_pos(rq);
220 sector_t block = sectors_to_logical(sdkp->device, sector);
221
222 if (!sd_is_zoned(sdkp))
223 /* Not a zoned device */
224 return BLK_STS_IOERR;
225
226 if (sdkp->device->changed)
227 return BLK_STS_IOERR;
228
229 if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
230 /* Unaligned request */
231 return BLK_STS_IOERR;
232
233 cmd->cmd_len = 16;
234 memset(cmd->cmnd, 0, cmd->cmd_len);
235 cmd->cmnd[0] = ZBC_OUT;
236 cmd->cmnd[1] = ZO_RESET_WRITE_POINTER;
237 put_unaligned_be64(block, &cmd->cmnd[2]);
238
239 rq->timeout = SD_TIMEOUT;
240 cmd->sc_data_direction = DMA_NONE;
241 cmd->transfersize = 0;
242 cmd->allowed = 0;
243
244 return BLK_STS_OK;
245 }
246
247 /**
248 * sd_zbc_complete - ZBC command post processing.
249 * @cmd: Completed command
250 * @good_bytes: Command reply bytes
251 * @sshdr: command sense header
252 *
253 * Called from sd_done(). Process report zones reply and handle reset zone
254 * and write commands errors.
255 */
256 void sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
257 struct scsi_sense_hdr *sshdr)
258 {
259 int result = cmd->result;
260 struct request *rq = cmd->request;
261
262 switch (req_op(rq)) {
263 case REQ_OP_ZONE_RESET:
264
265 if (result &&
266 sshdr->sense_key == ILLEGAL_REQUEST &&
267 sshdr->asc == 0x24)
268 /*
269 * INVALID FIELD IN CDB error: reset of a conventional
270 * zone was attempted. Nothing to worry about, so be
271 * quiet about the error.
272 */
273 rq->rq_flags |= RQF_QUIET;
274 break;
275
276 case REQ_OP_WRITE:
277 case REQ_OP_WRITE_ZEROES:
278 case REQ_OP_WRITE_SAME:
279 break;
280 }
281 }
282
283 /**
284 * sd_zbc_check_zoned_characteristics - Check zoned block device characteristics
285 * @sdkp: Target disk
286 * @buf: Buffer where to store the VPD page data
287 *
288 * Read VPD page B6, get information and check that reads are unconstrained.
289 */
290 static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
291 unsigned char *buf)
292 {
293
294 if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
295 sd_printk(KERN_NOTICE, sdkp,
296 "Read zoned characteristics VPD page failed\n");
297 return -ENODEV;
298 }
299
300 if (sdkp->device->type != TYPE_ZBC) {
301 /* Host-aware */
302 sdkp->urswrz = 1;
303 sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]);
304 sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]);
305 sdkp->zones_max_open = 0;
306 } else {
307 /* Host-managed */
308 sdkp->urswrz = buf[4] & 1;
309 sdkp->zones_optimal_open = 0;
310 sdkp->zones_optimal_nonseq = 0;
311 sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
312 }
313
314 /*
315 * Check for unconstrained reads: host-managed devices with
316 * constrained reads (drives failing read after write pointer)
317 * are not supported.
318 */
319 if (!sdkp->urswrz) {
320 if (sdkp->first_scan)
321 sd_printk(KERN_NOTICE, sdkp,
322 "constrained reads devices are not supported\n");
323 return -ENODEV;
324 }
325
326 return 0;
327 }
328
329 /**
330 * sd_zbc_check_zones - Check the device capacity and zone sizes
331 * @sdkp: Target disk
332 *
333 * Check that the device capacity as reported by READ CAPACITY matches the
334 * max_lba value (plus one)of the report zones command reply. Also check that
335 * all zones of the device have an equal size, only allowing the last zone of
336 * the disk to have a smaller size (runt zone). The zone size must also be a
337 * power of two.
338 *
339 * Returns the zone size in number of blocks upon success or an error code
340 * upon failure.
341 */
342 static int sd_zbc_check_zones(struct scsi_disk *sdkp, u32 *zblocks)
343 {
344 size_t bufsize, buflen;
345 unsigned int noio_flag;
346 u64 zone_blocks = 0;
347 sector_t max_lba, block = 0;
348 unsigned char *buf;
349 unsigned char *rec;
350 int ret;
351 u8 same;
352
353 /* Do all memory allocations as if GFP_NOIO was specified */
354 noio_flag = memalloc_noio_save();
355
356 /* Get a buffer */
357 buf = sd_zbc_alloc_report_buffer(sdkp, SD_ZBC_REPORT_MAX_ZONES,
358 &bufsize);
359 if (!buf) {
360 ret = -ENOMEM;
361 goto out;
362 }
363
364 /* Do a report zone to get max_lba and the same field */
365 ret = sd_zbc_do_report_zones(sdkp, buf, bufsize, 0, false);
366 if (ret)
367 goto out_free;
368
369 if (sdkp->rc_basis == 0) {
370 /* The max_lba field is the capacity of this device */
371 max_lba = get_unaligned_be64(&buf[8]);
372 if (sdkp->capacity != max_lba + 1) {
373 if (sdkp->first_scan)
374 sd_printk(KERN_WARNING, sdkp,
375 "Changing capacity from %llu to max LBA+1 %llu\n",
376 (unsigned long long)sdkp->capacity,
377 (unsigned long long)max_lba + 1);
378 sdkp->capacity = max_lba + 1;
379 }
380 }
381
382 /*
383 * Check same field: for any value other than 0, we know that all zones
384 * have the same size.
385 */
386 same = buf[4] & 0x0f;
387 if (same > 0) {
388 rec = &buf[64];
389 zone_blocks = get_unaligned_be64(&rec[8]);
390 goto out;
391 }
392
393 /*
394 * Check the size of all zones: all zones must be of
395 * equal size, except the last zone which can be smaller
396 * than other zones.
397 */
398 do {
399
400 /* Parse REPORT ZONES header */
401 buflen = min_t(size_t, get_unaligned_be32(&buf[0]) + 64,
402 bufsize);
403 rec = buf + 64;
404
405 /* Parse zone descriptors */
406 while (rec < buf + buflen) {
407 u64 this_zone_blocks = get_unaligned_be64(&rec[8]);
408
409 if (zone_blocks == 0) {
410 zone_blocks = this_zone_blocks;
411 } else if (this_zone_blocks != zone_blocks &&
412 (block + this_zone_blocks < sdkp->capacity
413 || this_zone_blocks > zone_blocks)) {
414 zone_blocks = 0;
415 goto out;
416 }
417 block += this_zone_blocks;
418 rec += 64;
419 }
420
421 if (block < sdkp->capacity) {
422 ret = sd_zbc_do_report_zones(sdkp, buf, bufsize, block,
423 true);
424 if (ret)
425 goto out_free;
426 }
427
428 } while (block < sdkp->capacity);
429
430 out:
431 if (!zone_blocks) {
432 if (sdkp->first_scan)
433 sd_printk(KERN_NOTICE, sdkp,
434 "Devices with non constant zone "
435 "size are not supported\n");
436 ret = -ENODEV;
437 } else if (!is_power_of_2(zone_blocks)) {
438 if (sdkp->first_scan)
439 sd_printk(KERN_NOTICE, sdkp,
440 "Devices with non power of 2 zone "
441 "size are not supported\n");
442 ret = -ENODEV;
443 } else if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
444 if (sdkp->first_scan)
445 sd_printk(KERN_NOTICE, sdkp,
446 "Zone size too large\n");
447 ret = -EFBIG;
448 } else {
449 *zblocks = zone_blocks;
450 ret = 0;
451 }
452
453 out_free:
454 memalloc_noio_restore(noio_flag);
455 kvfree(buf);
456
457 return ret;
458 }
459
460 int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
461 {
462 struct gendisk *disk = sdkp->disk;
463 unsigned int nr_zones;
464 u32 zone_blocks = 0;
465 int ret;
466
467 if (!sd_is_zoned(sdkp))
468 /*
469 * Device managed or normal SCSI disk,
470 * no special handling required
471 */
472 return 0;
473
474 /* Check zoned block device characteristics (unconstrained reads) */
475 ret = sd_zbc_check_zoned_characteristics(sdkp, buf);
476 if (ret)
477 goto err;
478
479 /*
480 * Check zone size: only devices with a constant zone size (except
481 * an eventual last runt zone) that is a power of 2 are supported.
482 */
483 ret = sd_zbc_check_zones(sdkp, &zone_blocks);
484 if (ret != 0)
485 goto err;
486
487 /* The drive satisfies the kernel restrictions: set it up */
488 blk_queue_chunk_sectors(sdkp->disk->queue,
489 logical_to_sectors(sdkp->device, zone_blocks));
490 nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
491
492 /* READ16/WRITE16 is mandatory for ZBC disks */
493 sdkp->device->use_16_for_rw = 1;
494 sdkp->device->use_10_for_rw = 0;
495
496 /*
497 * Revalidate the disk zone bitmaps once the block device capacity is
498 * set on the second revalidate execution during disk scan and if
499 * something changed when executing a normal revalidate.
500 */
501 if (sdkp->first_scan) {
502 sdkp->zone_blocks = zone_blocks;
503 sdkp->nr_zones = nr_zones;
504 return 0;
505 }
506
507 if (sdkp->zone_blocks != zone_blocks ||
508 sdkp->nr_zones != nr_zones ||
509 disk->queue->nr_zones != nr_zones) {
510 ret = blk_revalidate_disk_zones(disk);
511 if (ret != 0)
512 goto err;
513 sdkp->zone_blocks = zone_blocks;
514 sdkp->nr_zones = nr_zones;
515 }
516
517 return 0;
518
519 err:
520 sdkp->capacity = 0;
521
522 return ret;
523 }
524
525 void sd_zbc_print_zones(struct scsi_disk *sdkp)
526 {
527 if (!sd_is_zoned(sdkp) || !sdkp->capacity)
528 return;
529
530 if (sdkp->capacity & (sdkp->zone_blocks - 1))
531 sd_printk(KERN_NOTICE, sdkp,
532 "%u zones of %u logical blocks + 1 runt zone\n",
533 sdkp->nr_zones - 1,
534 sdkp->zone_blocks);
535 else
536 sd_printk(KERN_NOTICE, sdkp,
537 "%u zones of %u logical blocks\n",
538 sdkp->nr_zones,
539 sdkp->zone_blocks);
540 }