xref: /linux/drivers/scsi/sd_zbc.c (revision 2fe05e1139a555ae91f00a812cb9520e7d3022ab)
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 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_driver.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_eh.h>
36 
37 #include "sd.h"
38 #include "scsi_priv.h"
39 
40 enum zbc_zone_type {
41 	ZBC_ZONE_TYPE_CONV = 0x1,
42 	ZBC_ZONE_TYPE_SEQWRITE_REQ,
43 	ZBC_ZONE_TYPE_SEQWRITE_PREF,
44 	ZBC_ZONE_TYPE_RESERVED,
45 };
46 
47 enum zbc_zone_cond {
48 	ZBC_ZONE_COND_NO_WP,
49 	ZBC_ZONE_COND_EMPTY,
50 	ZBC_ZONE_COND_IMP_OPEN,
51 	ZBC_ZONE_COND_EXP_OPEN,
52 	ZBC_ZONE_COND_CLOSED,
53 	ZBC_ZONE_COND_READONLY = 0xd,
54 	ZBC_ZONE_COND_FULL,
55 	ZBC_ZONE_COND_OFFLINE,
56 };
57 
58 /**
59  * Convert a zone descriptor to a zone struct.
60  */
61 static void sd_zbc_parse_report(struct scsi_disk *sdkp,
62 				u8 *buf,
63 				struct blk_zone *zone)
64 {
65 	struct scsi_device *sdp = sdkp->device;
66 
67 	memset(zone, 0, sizeof(struct blk_zone));
68 
69 	zone->type = buf[0] & 0x0f;
70 	zone->cond = (buf[1] >> 4) & 0xf;
71 	if (buf[1] & 0x01)
72 		zone->reset = 1;
73 	if (buf[1] & 0x02)
74 		zone->non_seq = 1;
75 
76 	zone->len = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
77 	zone->start = logical_to_sectors(sdp, get_unaligned_be64(&buf[16]));
78 	zone->wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24]));
79 	if (zone->type != ZBC_ZONE_TYPE_CONV &&
80 	    zone->cond == ZBC_ZONE_COND_FULL)
81 		zone->wp = zone->start + zone->len;
82 }
83 
84 /**
85  * Issue a REPORT ZONES scsi command.
86  */
87 static int sd_zbc_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
88 			       unsigned int buflen, sector_t lba)
89 {
90 	struct scsi_device *sdp = sdkp->device;
91 	const int timeout = sdp->request_queue->rq_timeout;
92 	struct scsi_sense_hdr sshdr;
93 	unsigned char cmd[16];
94 	unsigned int rep_len;
95 	int result;
96 
97 	memset(cmd, 0, 16);
98 	cmd[0] = ZBC_IN;
99 	cmd[1] = ZI_REPORT_ZONES;
100 	put_unaligned_be64(lba, &cmd[2]);
101 	put_unaligned_be32(buflen, &cmd[10]);
102 	memset(buf, 0, buflen);
103 
104 	result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
105 				  buf, buflen, &sshdr,
106 				  timeout, SD_MAX_RETRIES, NULL);
107 	if (result) {
108 		sd_printk(KERN_ERR, sdkp,
109 			  "REPORT ZONES lba %llu failed with %d/%d\n",
110 			  (unsigned long long)lba,
111 			  host_byte(result), driver_byte(result));
112 		return -EIO;
113 	}
114 
115 	rep_len = get_unaligned_be32(&buf[0]);
116 	if (rep_len < 64) {
117 		sd_printk(KERN_ERR, sdkp,
118 			  "REPORT ZONES report invalid length %u\n",
119 			  rep_len);
120 		return -EIO;
121 	}
122 
123 	return 0;
124 }
125 
126 int sd_zbc_setup_report_cmnd(struct scsi_cmnd *cmd)
127 {
128 	struct request *rq = cmd->request;
129 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
130 	sector_t lba, sector = blk_rq_pos(rq);
131 	unsigned int nr_bytes = blk_rq_bytes(rq);
132 	int ret;
133 
134 	WARN_ON(nr_bytes == 0);
135 
136 	if (!sd_is_zoned(sdkp))
137 		/* Not a zoned device */
138 		return BLKPREP_KILL;
139 
140 	ret = scsi_init_io(cmd);
141 	if (ret != BLKPREP_OK)
142 		return ret;
143 
144 	cmd->cmd_len = 16;
145 	memset(cmd->cmnd, 0, cmd->cmd_len);
146 	cmd->cmnd[0] = ZBC_IN;
147 	cmd->cmnd[1] = ZI_REPORT_ZONES;
148 	lba = sectors_to_logical(sdkp->device, sector);
149 	put_unaligned_be64(lba, &cmd->cmnd[2]);
150 	put_unaligned_be32(nr_bytes, &cmd->cmnd[10]);
151 	/* Do partial report for speeding things up */
152 	cmd->cmnd[14] = ZBC_REPORT_ZONE_PARTIAL;
153 
154 	cmd->sc_data_direction = DMA_FROM_DEVICE;
155 	cmd->sdb.length = nr_bytes;
156 	cmd->transfersize = sdkp->device->sector_size;
157 	cmd->allowed = 0;
158 
159 	/*
160 	 * Report may return less bytes than requested. Make sure
161 	 * to report completion on the entire initial request.
162 	 */
163 	rq->__data_len = nr_bytes;
164 
165 	return BLKPREP_OK;
166 }
167 
168 static 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 static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
224 {
225 	return logical_to_sectors(sdkp->device, sdkp->zone_blocks);
226 }
227 
228 static inline unsigned int sd_zbc_zone_no(struct scsi_disk *sdkp,
229 					  sector_t sector)
230 {
231 	return sectors_to_logical(sdkp->device, sector) >> sdkp->zone_shift;
232 }
233 
234 int sd_zbc_setup_reset_cmnd(struct scsi_cmnd *cmd)
235 {
236 	struct request *rq = cmd->request;
237 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
238 	sector_t sector = blk_rq_pos(rq);
239 	sector_t block = sectors_to_logical(sdkp->device, sector);
240 
241 	if (!sd_is_zoned(sdkp))
242 		/* Not a zoned device */
243 		return BLKPREP_KILL;
244 
245 	if (sdkp->device->changed)
246 		return BLKPREP_KILL;
247 
248 	if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
249 		/* Unaligned request */
250 		return BLKPREP_KILL;
251 
252 	cmd->cmd_len = 16;
253 	memset(cmd->cmnd, 0, cmd->cmd_len);
254 	cmd->cmnd[0] = ZBC_OUT;
255 	cmd->cmnd[1] = ZO_RESET_WRITE_POINTER;
256 	put_unaligned_be64(block, &cmd->cmnd[2]);
257 
258 	rq->timeout = SD_TIMEOUT;
259 	cmd->sc_data_direction = DMA_NONE;
260 	cmd->transfersize = 0;
261 	cmd->allowed = 0;
262 
263 	return BLKPREP_OK;
264 }
265 
266 int sd_zbc_write_lock_zone(struct scsi_cmnd *cmd)
267 {
268 	struct request *rq = cmd->request;
269 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
270 	sector_t sector = blk_rq_pos(rq);
271 	sector_t zone_sectors = sd_zbc_zone_sectors(sdkp);
272 	unsigned int zno = sd_zbc_zone_no(sdkp, sector);
273 
274 	/*
275 	 * Note: Checks of the alignment of the write command on
276 	 * logical blocks is done in sd.c
277 	 */
278 
279 	/* Do not allow zone boundaries crossing on host-managed drives */
280 	if (blk_queue_zoned_model(sdkp->disk->queue) == BLK_ZONED_HM &&
281 	    (sector & (zone_sectors - 1)) + blk_rq_sectors(rq) > zone_sectors)
282 		return BLKPREP_KILL;
283 
284 	/*
285 	 * Do not issue more than one write at a time per
286 	 * zone. This solves write ordering problems due to
287 	 * the unlocking of the request queue in the dispatch
288 	 * path in the non scsi-mq case. For scsi-mq, this
289 	 * also avoids potential write reordering when multiple
290 	 * threads running on different CPUs write to the same
291 	 * zone (with a synchronized sequential pattern).
292 	 */
293 	if (sdkp->zones_wlock &&
294 	    test_and_set_bit(zno, sdkp->zones_wlock))
295 		return BLKPREP_DEFER;
296 
297 	return BLKPREP_OK;
298 }
299 
300 void sd_zbc_write_unlock_zone(struct scsi_cmnd *cmd)
301 {
302 	struct request *rq = cmd->request;
303 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
304 
305 	if (sdkp->zones_wlock) {
306 		unsigned int zno = sd_zbc_zone_no(sdkp, blk_rq_pos(rq));
307 		WARN_ON_ONCE(!test_bit(zno, sdkp->zones_wlock));
308 		clear_bit_unlock(zno, sdkp->zones_wlock);
309 		smp_mb__after_atomic();
310 	}
311 }
312 
313 void sd_zbc_complete(struct scsi_cmnd *cmd,
314 		     unsigned int good_bytes,
315 		     struct scsi_sense_hdr *sshdr)
316 {
317 	int result = cmd->result;
318 	struct request *rq = cmd->request;
319 
320 	switch (req_op(rq)) {
321 	case REQ_OP_ZONE_RESET:
322 
323 		if (result &&
324 		    sshdr->sense_key == ILLEGAL_REQUEST &&
325 		    sshdr->asc == 0x24)
326 			/*
327 			 * INVALID FIELD IN CDB error: reset of a conventional
328 			 * zone was attempted. Nothing to worry about, so be
329 			 * quiet about the error.
330 			 */
331 			rq->rq_flags |= RQF_QUIET;
332 		break;
333 
334 	case REQ_OP_WRITE:
335 	case REQ_OP_WRITE_ZEROES:
336 	case REQ_OP_WRITE_SAME:
337 
338 		/* Unlock the zone */
339 		sd_zbc_write_unlock_zone(cmd);
340 
341 		if (result &&
342 		    sshdr->sense_key == ILLEGAL_REQUEST &&
343 		    sshdr->asc == 0x21)
344 			/*
345 			 * INVALID ADDRESS FOR WRITE error: It is unlikely that
346 			 * retrying write requests failed with any kind of
347 			 * alignement error will result in success. So don't.
348 			 */
349 			cmd->allowed = 0;
350 		break;
351 
352 	case REQ_OP_ZONE_REPORT:
353 
354 		if (!result)
355 			sd_zbc_report_zones_complete(cmd, good_bytes);
356 		break;
357 
358 	}
359 }
360 
361 /**
362  * Read zoned block device characteristics (VPD page B6).
363  */
364 static int sd_zbc_read_zoned_characteristics(struct scsi_disk *sdkp,
365 					     unsigned char *buf)
366 {
367 
368 	if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
369 		sd_printk(KERN_NOTICE, sdkp,
370 			  "Unconstrained-read check failed\n");
371 		return -ENODEV;
372 	}
373 
374 	if (sdkp->device->type != TYPE_ZBC) {
375 		/* Host-aware */
376 		sdkp->urswrz = 1;
377 		sdkp->zones_optimal_open = get_unaligned_be64(&buf[8]);
378 		sdkp->zones_optimal_nonseq = get_unaligned_be64(&buf[12]);
379 		sdkp->zones_max_open = 0;
380 	} else {
381 		/* Host-managed */
382 		sdkp->urswrz = buf[4] & 1;
383 		sdkp->zones_optimal_open = 0;
384 		sdkp->zones_optimal_nonseq = 0;
385 		sdkp->zones_max_open = get_unaligned_be64(&buf[16]);
386 	}
387 
388 	return 0;
389 }
390 
391 /**
392  * Check reported capacity.
393  */
394 static int sd_zbc_check_capacity(struct scsi_disk *sdkp,
395 				 unsigned char *buf)
396 {
397 	sector_t lba;
398 	int ret;
399 
400 	if (sdkp->rc_basis != 0)
401 		return 0;
402 
403 	/* Do a report zone to get the maximum LBA to check capacity */
404 	ret = sd_zbc_report_zones(sdkp, buf, SD_BUF_SIZE, 0);
405 	if (ret)
406 		return ret;
407 
408 	/* The max_lba field is the capacity of this device */
409 	lba = get_unaligned_be64(&buf[8]);
410 	if (lba + 1 == sdkp->capacity)
411 		return 0;
412 
413 	if (sdkp->first_scan)
414 		sd_printk(KERN_WARNING, sdkp,
415 			  "Changing capacity from %llu to max LBA+1 %llu\n",
416 			  (unsigned long long)sdkp->capacity,
417 			  (unsigned long long)lba + 1);
418 	sdkp->capacity = lba + 1;
419 
420 	return 0;
421 }
422 
423 #define SD_ZBC_BUF_SIZE 131072
424 
425 static int sd_zbc_check_zone_size(struct scsi_disk *sdkp)
426 {
427 	u64 zone_blocks;
428 	sector_t block = 0;
429 	unsigned char *buf;
430 	unsigned char *rec;
431 	unsigned int buf_len;
432 	unsigned int list_length;
433 	int ret;
434 	u8 same;
435 
436 	sdkp->zone_blocks = 0;
437 
438 	/* Get a buffer */
439 	buf = kmalloc(SD_ZBC_BUF_SIZE, GFP_KERNEL);
440 	if (!buf)
441 		return -ENOMEM;
442 
443 	/* Do a report zone to get the same field */
444 	ret = sd_zbc_report_zones(sdkp, buf, SD_ZBC_BUF_SIZE, 0);
445 	if (ret) {
446 		zone_blocks = 0;
447 		goto out;
448 	}
449 
450 	same = buf[4] & 0x0f;
451 	if (same > 0) {
452 		rec = &buf[64];
453 		zone_blocks = get_unaligned_be64(&rec[8]);
454 		goto out;
455 	}
456 
457 	/*
458 	 * Check the size of all zones: all zones must be of
459 	 * equal size, except the last zone which can be smaller
460 	 * than other zones.
461 	 */
462 	do {
463 
464 		/* Parse REPORT ZONES header */
465 		list_length = get_unaligned_be32(&buf[0]) + 64;
466 		rec = buf + 64;
467 		if (list_length < SD_ZBC_BUF_SIZE)
468 			buf_len = list_length;
469 		else
470 			buf_len = SD_ZBC_BUF_SIZE;
471 
472 		/* Parse zone descriptors */
473 		while (rec < buf + buf_len) {
474 			zone_blocks = get_unaligned_be64(&rec[8]);
475 			if (sdkp->zone_blocks == 0) {
476 				sdkp->zone_blocks = zone_blocks;
477 			} else if (zone_blocks != sdkp->zone_blocks &&
478 				   (block + zone_blocks < sdkp->capacity
479 				    || zone_blocks > sdkp->zone_blocks)) {
480 				zone_blocks = 0;
481 				goto out;
482 			}
483 			block += zone_blocks;
484 			rec += 64;
485 		}
486 
487 		if (block < sdkp->capacity) {
488 			ret = sd_zbc_report_zones(sdkp, buf,
489 						  SD_ZBC_BUF_SIZE, block);
490 			if (ret)
491 				return ret;
492 		}
493 
494 	} while (block < sdkp->capacity);
495 
496 	zone_blocks = sdkp->zone_blocks;
497 
498 out:
499 	kfree(buf);
500 
501 	if (!zone_blocks) {
502 		if (sdkp->first_scan)
503 			sd_printk(KERN_NOTICE, sdkp,
504 				  "Devices with non constant zone "
505 				  "size are not supported\n");
506 		return -ENODEV;
507 	}
508 
509 	if (!is_power_of_2(zone_blocks)) {
510 		if (sdkp->first_scan)
511 			sd_printk(KERN_NOTICE, sdkp,
512 				  "Devices with non power of 2 zone "
513 				  "size are not supported\n");
514 		return -ENODEV;
515 	}
516 
517 	if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
518 		if (sdkp->first_scan)
519 			sd_printk(KERN_NOTICE, sdkp,
520 				  "Zone size too large\n");
521 		return -ENODEV;
522 	}
523 
524 	sdkp->zone_blocks = zone_blocks;
525 
526 	return 0;
527 }
528 
529 static int sd_zbc_setup(struct scsi_disk *sdkp)
530 {
531 
532 	/* chunk_sectors indicates the zone size */
533 	blk_queue_chunk_sectors(sdkp->disk->queue,
534 			logical_to_sectors(sdkp->device, sdkp->zone_blocks));
535 	sdkp->zone_shift = ilog2(sdkp->zone_blocks);
536 	sdkp->nr_zones = sdkp->capacity >> sdkp->zone_shift;
537 	if (sdkp->capacity & (sdkp->zone_blocks - 1))
538 		sdkp->nr_zones++;
539 
540 	if (!sdkp->zones_wlock) {
541 		sdkp->zones_wlock = kcalloc(BITS_TO_LONGS(sdkp->nr_zones),
542 					    sizeof(unsigned long),
543 					    GFP_KERNEL);
544 		if (!sdkp->zones_wlock)
545 			return -ENOMEM;
546 	}
547 
548 	return 0;
549 }
550 
551 int sd_zbc_read_zones(struct scsi_disk *sdkp,
552 		      unsigned char *buf)
553 {
554 	int ret;
555 
556 	if (!sd_is_zoned(sdkp))
557 		/*
558 		 * Device managed or normal SCSI disk,
559 		 * no special handling required
560 		 */
561 		return 0;
562 
563 
564 	/* Get zoned block device characteristics */
565 	ret = sd_zbc_read_zoned_characteristics(sdkp, buf);
566 	if (ret)
567 		goto err;
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 		ret = -ENODEV;
579 		goto err;
580 	}
581 
582 	/* Check capacity */
583 	ret = sd_zbc_check_capacity(sdkp, buf);
584 	if (ret)
585 		goto err;
586 
587 	/*
588 	 * Check zone size: only devices with a constant zone size (except
589 	 * an eventual last runt zone) that is a power of 2 are supported.
590 	 */
591 	ret = sd_zbc_check_zone_size(sdkp);
592 	if (ret)
593 		goto err;
594 
595 	/* The drive satisfies the kernel restrictions: set it up */
596 	ret = sd_zbc_setup(sdkp);
597 	if (ret)
598 		goto err;
599 
600 	/* READ16/WRITE16 is mandatory for ZBC disks */
601 	sdkp->device->use_16_for_rw = 1;
602 	sdkp->device->use_10_for_rw = 0;
603 
604 	return 0;
605 
606 err:
607 	sdkp->capacity = 0;
608 
609 	return ret;
610 }
611 
612 void sd_zbc_remove(struct scsi_disk *sdkp)
613 {
614 	kfree(sdkp->zones_wlock);
615 	sdkp->zones_wlock = NULL;
616 }
617 
618 void sd_zbc_print_zones(struct scsi_disk *sdkp)
619 {
620 	if (!sd_is_zoned(sdkp) || !sdkp->capacity)
621 		return;
622 
623 	if (sdkp->capacity & (sdkp->zone_blocks - 1))
624 		sd_printk(KERN_NOTICE, sdkp,
625 			  "%u zones of %u logical blocks + 1 runt zone\n",
626 			  sdkp->nr_zones - 1,
627 			  sdkp->zone_blocks);
628 	else
629 		sd_printk(KERN_NOTICE, sdkp,
630 			  "%u zones of %u logical blocks\n",
631 			  sdkp->nr_zones,
632 			  sdkp->zone_blocks);
633 }
634