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#include <linux/mutex.h>
 15
 16#include <linux/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/* Whether or not a SCSI zone descriptor describes a gap zone. */
 27static bool sd_zbc_is_gap_zone(const u8 buf[64])
 28{
 29	return (buf[0] & 0xf) == ZBC_ZONE_TYPE_GAP;
 30}
 31
 32/**
 33 * sd_zbc_parse_report - Parse a SCSI zone descriptor
 34 * @sdkp: SCSI disk pointer.
 35 * @buf: SCSI zone descriptor.
 36 * @idx: Index of the zone relative to the first zone reported by the current
 37 *	sd_zbc_report_zones() call.
 38 * @args: report zones arguments (callback, etc)
 39 *
 40 * Return: Value returned by @cb.
 41 *
 42 * Convert a SCSI zone descriptor into struct blk_zone format. Additionally,
 43 * call @cb(blk_zone, @data).
 44 */
 45static int sd_zbc_parse_report(struct scsi_disk *sdkp, const u8 buf[64],
 46			unsigned int idx, struct blk_report_zones_args *args)
 47{
 48	struct scsi_device *sdp = sdkp->device;
 49	struct blk_zone zone = { 0 };
 50	sector_t start_lba, gran;
 51
 52	if (WARN_ON_ONCE(sd_zbc_is_gap_zone(buf)))
 53		return -EINVAL;
 54
 55	zone.type = buf[0] & 0x0f;
 56	zone.cond = (buf[1] >> 4) & 0xf;
 57	if (buf[1] & 0x01)
 58		zone.reset = 1;
 59	if (buf[1] & 0x02)
 60		zone.non_seq = 1;
 61
 62	start_lba = get_unaligned_be64(&buf[16]);
 63	zone.start = logical_to_sectors(sdp, start_lba);
 64	zone.capacity = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
 65	zone.len = zone.capacity;
 66	if (sdkp->zone_starting_lba_gran) {
 67		gran = logical_to_sectors(sdp, sdkp->zone_starting_lba_gran);
 68		if (zone.len > gran) {
 69			sd_printk(KERN_ERR, sdkp,
 70				  "Invalid zone at LBA %llu with capacity %llu and length %llu; granularity = %llu\n",
 71				  start_lba,
 72				  sectors_to_logical(sdp, zone.capacity),
 73				  sectors_to_logical(sdp, zone.len),
 74				  sectors_to_logical(sdp, gran));
 75			return -EINVAL;
 76		}
 77		/*
 78		 * Use the starting LBA granularity instead of the zone length
 79		 * obtained from the REPORT ZONES command.
 80		 */
 81		zone.len = gran;
 82	}
 83	if (zone.cond == ZBC_ZONE_COND_FULL)
 84		zone.wp = zone.start + zone.len;
 85	else
 86		zone.wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24]));
 87
 88	return disk_report_zone(sdkp->disk, &zone, idx, args);
 89}
 90
 91/**
 92 * sd_zbc_do_report_zones - Issue a REPORT ZONES scsi command.
 93 * @sdkp: The target disk
 94 * @buf: vmalloc-ed buffer to use for the reply
 95 * @buflen: the buffer size
 96 * @lba: Start LBA of the report
 97 * @partial: Do partial report
 98 *
 99 * For internal use during device validation.
100 * Using partial=true can significantly speed up execution of a report zones
101 * command because the disk does not have to count all possible report matching
102 * zones and will only report the count of zones fitting in the command reply
103 * buffer.
104 */
105static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
106				  unsigned int buflen, sector_t lba,
107				  bool partial)
108{
109	struct scsi_device *sdp = sdkp->device;
110	const int timeout = sdp->request_queue->rq_timeout;
111	struct scsi_sense_hdr sshdr;
112	const struct scsi_exec_args exec_args = {
113		.sshdr = &sshdr,
114	};
115	unsigned char cmd[16];
116	unsigned int rep_len;
117	int result;
118
119	memset(cmd, 0, 16);
120	cmd[0] = ZBC_IN;
121	cmd[1] = ZI_REPORT_ZONES;
122	put_unaligned_be64(lba, &cmd[2]);
123	put_unaligned_be32(buflen, &cmd[10]);
124	if (partial)
125		cmd[14] = ZBC_REPORT_ZONE_PARTIAL;
126
127	result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buf, buflen,
128				  timeout, SD_MAX_RETRIES, &exec_args);
129	if (result) {
130		sd_printk(KERN_ERR, sdkp,
131			  "REPORT ZONES start lba %llu failed\n", lba);
132		sd_print_result(sdkp, "REPORT ZONES", result);
133		if (result > 0 && scsi_sense_valid(&sshdr))
134			sd_print_sense_hdr(sdkp, &sshdr);
135		return -EIO;
136	}
137
138	rep_len = get_unaligned_be32(&buf[0]);
139	if (rep_len < 64) {
140		sd_printk(KERN_ERR, sdkp,
141			  "REPORT ZONES report invalid length %u\n",
142			  rep_len);
143		return -EIO;
144	}
145
146	return 0;
147}
148
149/**
150 * sd_zbc_alloc_report_buffer() - Allocate a buffer for report zones reply.
151 * @sdkp: The target disk
152 * @nr_zones: Maximum number of zones to report
153 * @buflen: Size of the buffer allocated
154 *
155 * Try to allocate a reply buffer for the number of requested zones.
156 * The size of the buffer allocated may be smaller than requested to
157 * satify the device constraint (max_hw_sectors, max_segments, etc).
158 *
159 * Return the address of the allocated buffer and update @buflen with
160 * the size of the allocated buffer.
161 */
162static void *sd_zbc_alloc_report_buffer(struct scsi_disk *sdkp,
163					unsigned int nr_zones, size_t *buflen)
164{
165	struct request_queue *q = sdkp->disk->queue;
166	unsigned int max_segments;
167	size_t bufsize;
168	void *buf;
169
170	/*
171	 * Report zone buffer size should be at most 64B times the number of
172	 * zones requested plus the 64B reply header, but should be aligned
173	 * to SECTOR_SIZE for ATA devices.
174	 * Make sure that this size does not exceed the hardware capabilities.
175	 * Furthermore, since the report zone command cannot be split, make
176	 * sure that the allocated buffer can always be mapped by limiting the
177	 * number of pages allocated to the HBA max segments limit.
178	 * Since max segments can be larger than the max inline bio vectors,
179	 * further limit the allocated buffer to BIO_MAX_INLINE_VECS.
180	 */
181	nr_zones = min(nr_zones, sdkp->zone_info.nr_zones);
182	bufsize = roundup((nr_zones + 1) * 64, SECTOR_SIZE);
183	bufsize = min_t(size_t, bufsize,
184			queue_max_hw_sectors(q) << SECTOR_SHIFT);
185	max_segments = min(BIO_MAX_INLINE_VECS, queue_max_segments(q));
186	bufsize = min_t(size_t, bufsize, max_segments << PAGE_SHIFT);
187
188	while (bufsize >= SECTOR_SIZE) {
189		buf = kvzalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
190		if (buf) {
191			*buflen = bufsize;
192			return buf;
193		}
194		bufsize = rounddown(bufsize >> 1, SECTOR_SIZE);
195	}
196
197	return NULL;
198}
199
200/**
201 * sd_zbc_zone_sectors - Get the device zone size in number of 512B sectors.
202 * @sdkp: The target disk
203 */
204static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
205{
206	return logical_to_sectors(sdkp->device, sdkp->zone_info.zone_blocks);
207}
208
209/**
210 * sd_zbc_report_zones - SCSI .report_zones() callback.
211 * @disk: Disk to report zones for.
212 * @sector: Start sector.
213 * @nr_zones: Maximum number of zones to report.
214 * @args: Callback arguments.
215 *
216 * Called by the block layer to iterate over zone information. See also the
217 * disk->fops->report_zones() calls in block/blk-zoned.c.
218 */
219int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
220			unsigned int nr_zones,
221			struct blk_report_zones_args *args)
222{
223	struct scsi_disk *sdkp = scsi_disk(disk);
224	sector_t lba = sectors_to_logical(sdkp->device, sector);
225	unsigned int nr, i;
226	unsigned char *buf;
227	u64 zone_length, start_lba;
228	size_t offset, buflen = 0;
229	int zone_idx = 0;
230	int ret;
231
232	if (sdkp->device->type != TYPE_ZBC)
233		/* Not a zoned device */
234		return -EOPNOTSUPP;
235
236	if (!sdkp->capacity)
237		/* Device gone or invalid */
238		return -ENODEV;
239
240	buf = sd_zbc_alloc_report_buffer(sdkp, nr_zones, &buflen);
241	if (!buf)
242		return -ENOMEM;
243
244	while (zone_idx < nr_zones && lba < sdkp->capacity) {
245		ret = sd_zbc_do_report_zones(sdkp, buf, buflen, lba, true);
246		if (ret)
247			goto out;
248
249		offset = 0;
250		nr = min(nr_zones, get_unaligned_be32(&buf[0]) / 64);
251		if (!nr)
252			break;
253
254		for (i = 0; i < nr && zone_idx < nr_zones; i++) {
255			offset += 64;
256			start_lba = get_unaligned_be64(&buf[offset + 16]);
257			zone_length = get_unaligned_be64(&buf[offset + 8]);
258			if ((zone_idx == 0 &&
259			    (lba < start_lba ||
260			     lba >= start_lba + zone_length)) ||
261			    (zone_idx > 0 && start_lba != lba) ||
262			    start_lba + zone_length < start_lba) {
263				sd_printk(KERN_ERR, sdkp,
264					  "Zone %d at LBA %llu is invalid: %llu + %llu\n",
265					  zone_idx, lba, start_lba, zone_length);
266				ret = -EINVAL;
267				goto out;
268			}
269			lba = start_lba + zone_length;
270			if (sd_zbc_is_gap_zone(&buf[offset])) {
271				if (sdkp->zone_starting_lba_gran)
272					continue;
273				sd_printk(KERN_ERR, sdkp,
274					  "Gap zone without constant LBA offsets\n");
275				ret = -EINVAL;
276				goto out;
277			}
278
279			ret = sd_zbc_parse_report(sdkp, buf + offset, zone_idx,
280						  args);
281			if (ret)
282				goto out;
283
284			zone_idx++;
285		}
286	}
287
288	ret = zone_idx;
289out:
290	kvfree(buf);
291	return ret;
292}
293
294static blk_status_t sd_zbc_cmnd_checks(struct scsi_cmnd *cmd)
295{
296	struct request *rq = scsi_cmd_to_rq(cmd);
297	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
298	sector_t sector = blk_rq_pos(rq);
299
300	if (sdkp->device->type != TYPE_ZBC)
301		/* Not a zoned device */
302		return BLK_STS_IOERR;
303
304	if (sdkp->device->changed)
305		return BLK_STS_IOERR;
306
307	if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
308		/* Unaligned request */
309		return BLK_STS_IOERR;
310
311	return BLK_STS_OK;
312}
313
314/**
315 * sd_zbc_setup_zone_mgmt_cmnd - Prepare a zone ZBC_OUT command. The operations
316 *			can be RESET WRITE POINTER, OPEN, CLOSE or FINISH.
317 * @cmd: the command to setup
318 * @op: Operation to be performed
319 * @all: All zones control
320 *
321 * Called from sd_init_command() for REQ_OP_ZONE_RESET, REQ_OP_ZONE_RESET_ALL,
322 * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE or REQ_OP_ZONE_FINISH requests.
323 */
324blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
325					 unsigned char op, bool all)
326{
327	struct request *rq = scsi_cmd_to_rq(cmd);
328	sector_t sector = blk_rq_pos(rq);
329	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
330	sector_t block = sectors_to_logical(sdkp->device, sector);
331	blk_status_t ret;
332
333	ret = sd_zbc_cmnd_checks(cmd);
334	if (ret != BLK_STS_OK)
335		return ret;
336
337	cmd->cmd_len = 16;
338	memset(cmd->cmnd, 0, cmd->cmd_len);
339	cmd->cmnd[0] = ZBC_OUT;
340	cmd->cmnd[1] = op;
341	if (all)
342		cmd->cmnd[14] = 0x1;
343	else
344		put_unaligned_be64(block, &cmd->cmnd[2]);
345
346	rq->timeout = SD_TIMEOUT;
347	cmd->sc_data_direction = DMA_NONE;
348	cmd->transfersize = 0;
349	cmd->allowed = 0;
350
351	return BLK_STS_OK;
352}
353
354/**
355 * sd_zbc_complete - ZBC command post processing.
356 * @cmd: Completed command
357 * @good_bytes: Command reply bytes
358 * @sshdr: command sense header
359 *
360 * Called from sd_done() to handle zone commands errors and updates to the
361 * device queue zone write pointer offset cahce.
362 */
363unsigned int sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
364		     struct scsi_sense_hdr *sshdr)
365{
366	int result = cmd->result;
367	struct request *rq = scsi_cmd_to_rq(cmd);
368
369	if (op_is_zone_mgmt(req_op(rq)) &&
370	    result &&
371	    sshdr->sense_key == ILLEGAL_REQUEST &&
372	    sshdr->asc == 0x24) {
373		/*
374		 * INVALID FIELD IN CDB error: a zone management command was
375		 * attempted on a conventional zone. Nothing to worry about,
376		 * so be quiet about the error.
377		 */
378		rq->rq_flags |= RQF_QUIET;
379	}
380
381	return good_bytes;
382}
383
384/**
385 * sd_zbc_check_zoned_characteristics - Check zoned block device characteristics
386 * @sdkp: Target disk
387 * @buf: Buffer where to store the VPD page data
388 *
389 * Read VPD page B6, get information and check that reads are unconstrained.
390 */
391static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
392					      unsigned char *buf)
393{
394	u64 zone_starting_lba_gran;
395
396	if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
397		sd_printk(KERN_NOTICE, sdkp,
398			  "Read zoned characteristics VPD page failed\n");
399		return -ENODEV;
400	}
401
402	if (sdkp->device->type != TYPE_ZBC) {
403		/* Host-aware */
404		sdkp->urswrz = 1;
405		sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]);
406		sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]);
407		sdkp->zones_max_open = 0;
408		return 0;
409	}
410
411	/* Host-managed */
412	sdkp->urswrz = buf[4] & 1;
413	sdkp->zones_optimal_open = 0;
414	sdkp->zones_optimal_nonseq = 0;
415	sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
416	/* Check zone alignment method */
417	switch (buf[23] & 0xf) {
418	case 0:
419	case ZBC_CONSTANT_ZONE_LENGTH:
420		/* Use zone length */
421		break;
422	case ZBC_CONSTANT_ZONE_START_OFFSET:
423		zone_starting_lba_gran = get_unaligned_be64(&buf[24]);
424		if (zone_starting_lba_gran == 0 ||
425		    !is_power_of_2(zone_starting_lba_gran) ||
426		    logical_to_sectors(sdkp->device, zone_starting_lba_gran) >
427		    UINT_MAX) {
428			sd_printk(KERN_ERR, sdkp,
429				  "Invalid zone starting LBA granularity %llu\n",
430				  zone_starting_lba_gran);
431			return -ENODEV;
432		}
433		sdkp->zone_starting_lba_gran = zone_starting_lba_gran;
434		break;
435	default:
436		sd_printk(KERN_ERR, sdkp, "Invalid zone alignment method\n");
437		return -ENODEV;
438	}
439
440	/*
441	 * Check for unconstrained reads: host-managed devices with
442	 * constrained reads (drives failing read after write pointer)
443	 * are not supported.
444	 */
445	if (!sdkp->urswrz) {
446		if (sdkp->first_scan)
447			sd_printk(KERN_NOTICE, sdkp,
448			  "constrained reads devices are not supported\n");
449		return -ENODEV;
450	}
451
452	return 0;
453}
454
455/**
456 * sd_zbc_check_capacity - Check the device capacity
457 * @sdkp: Target disk
458 * @buf: command buffer
459 * @zblocks: zone size in logical blocks
460 *
461 * Get the device zone size and check that the device capacity as reported
462 * by READ CAPACITY matches the max_lba value (plus one) of the report zones
463 * command reply for devices with RC_BASIS == 0.
464 *
465 * Returns 0 upon success or an error code upon failure.
466 */
467static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf,
468				 u32 *zblocks)
469{
470	u64 zone_blocks;
471	sector_t max_lba;
472	unsigned char *rec;
473	int ret;
474
475	/* Do a report zone to get max_lba and the size of the first zone */
476	ret = sd_zbc_do_report_zones(sdkp, buf, SD_BUF_SIZE, 0, false);
477	if (ret)
478		return ret;
479
480	if (sdkp->rc_basis == 0) {
481		/* The max_lba field is the capacity of this device */
482		max_lba = get_unaligned_be64(&buf[8]);
483		if (sdkp->capacity != max_lba + 1) {
484			if (sdkp->first_scan)
485				sd_printk(KERN_WARNING, sdkp,
486					"Changing capacity from %llu to max LBA+1 %llu\n",
487					(unsigned long long)sdkp->capacity,
488					(unsigned long long)max_lba + 1);
489			sdkp->capacity = max_lba + 1;
490		}
491	}
492
493	if (sdkp->zone_starting_lba_gran == 0) {
494		/* Get the size of the first reported zone */
495		rec = buf + 64;
496		zone_blocks = get_unaligned_be64(&rec[8]);
497		if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
498			if (sdkp->first_scan)
499				sd_printk(KERN_NOTICE, sdkp,
500					  "Zone size too large\n");
501			return -EFBIG;
502		}
503	} else {
504		zone_blocks = sdkp->zone_starting_lba_gran;
505	}
506
507	if (!is_power_of_2(zone_blocks)) {
508		sd_printk(KERN_ERR, sdkp,
509			  "Zone size %llu is not a power of two.\n",
510			  zone_blocks);
511		return -EINVAL;
512	}
513
514	*zblocks = zone_blocks;
515
516	return 0;
517}
518
519static void sd_zbc_print_zones(struct scsi_disk *sdkp)
520{
521	if (sdkp->device->type != TYPE_ZBC || !sdkp->capacity)
522		return;
523
524	if (sdkp->capacity & (sdkp->zone_info.zone_blocks - 1))
525		sd_printk(KERN_NOTICE, sdkp,
526			  "%u zones of %u logical blocks + 1 runt zone\n",
527			  sdkp->zone_info.nr_zones - 1,
528			  sdkp->zone_info.zone_blocks);
529	else
530		sd_printk(KERN_NOTICE, sdkp,
531			  "%u zones of %u logical blocks\n",
532			  sdkp->zone_info.nr_zones,
533			  sdkp->zone_info.zone_blocks);
534}
535
536/*
537 * Call blk_revalidate_disk_zones() if any of the zoned disk properties have
538 * changed that make it necessary to call that function. Called by
539 * sd_revalidate_disk() after the gendisk capacity has been set.
540 */
541int sd_zbc_revalidate_zones(struct scsi_disk *sdkp)
542{
543	struct gendisk *disk = sdkp->disk;
544	struct request_queue *q = disk->queue;
545	u32 zone_blocks = sdkp->early_zone_info.zone_blocks;
546	unsigned int nr_zones = sdkp->early_zone_info.nr_zones;
547	unsigned int flags;
548	int ret;
549
550	/*
551	 * There is nothing to do for regular disks, including host-aware disks
552	 * that have partitions.
553	 */
554	if (!blk_queue_is_zoned(q))
555		return 0;
556
557	if (sdkp->zone_info.zone_blocks == zone_blocks &&
558	    sdkp->zone_info.nr_zones == nr_zones &&
559	    disk->nr_zones == nr_zones)
560		return 0;
561
562	sdkp->zone_info.zone_blocks = zone_blocks;
563	sdkp->zone_info.nr_zones = nr_zones;
564
565	flags = memalloc_noio_save();
566	ret = blk_revalidate_disk_zones(disk);
567	memalloc_noio_restore(flags);
568	if (ret) {
569		sdkp->zone_info = (struct zoned_disk_info){ };
570		sdkp->capacity = 0;
571		return ret;
572	}
573
574	sd_zbc_print_zones(sdkp);
575
576	return 0;
577}
578
579/**
580 * sd_zbc_read_zones - Read zone information and update the request queue
581 * @sdkp: SCSI disk pointer.
582 * @lim: queue limits to read into
583 * @buf: 512 byte buffer used for storing SCSI command output.
584 *
585 * Read zone information and update the request queue zone characteristics and
586 * also the zoned device information in *sdkp. Called by sd_revalidate_disk()
587 * before the gendisk capacity has been set.
588 */
589int sd_zbc_read_zones(struct scsi_disk *sdkp, struct queue_limits *lim,
590		u8 buf[SD_BUF_SIZE])
591{
592	unsigned int nr_zones;
593	u32 zone_blocks = 0;
594	int ret;
595
596	if (sdkp->device->type != TYPE_ZBC)
597		return 0;
598
599	lim->features |= BLK_FEAT_ZONED;
600
601	/*
602	 * Per ZBC and ZAC specifications, writes in sequential write required
603	 * zones of host-managed devices must be aligned to the device physical
604	 * block size.
605	 */
606	lim->zone_write_granularity = sdkp->physical_block_size;
607
608	/* READ16/WRITE16/SYNC16 is mandatory for ZBC devices */
609	sdkp->device->use_16_for_rw = 1;
610	sdkp->device->use_10_for_rw = 0;
611	sdkp->device->use_16_for_sync = 1;
612
613	/* Check zoned block device characteristics (unconstrained reads) */
614	ret = sd_zbc_check_zoned_characteristics(sdkp, buf);
615	if (ret)
616		goto err;
617
618	/* Check the device capacity reported by report zones */
619	ret = sd_zbc_check_capacity(sdkp, buf, &zone_blocks);
620	if (ret != 0)
621		goto err;
622
623	nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
624	sdkp->early_zone_info.nr_zones = nr_zones;
625	sdkp->early_zone_info.zone_blocks = zone_blocks;
626
627	/* The drive satisfies the kernel restrictions: set it up */
628	if (sdkp->zones_max_open == U32_MAX)
629		lim->max_open_zones = 0;
630	else
631		lim->max_open_zones = sdkp->zones_max_open;
632	lim->max_active_zones = 0;
633	lim->chunk_sectors = logical_to_sectors(sdkp->device, zone_blocks);
634
635	return 0;
636
637err:
638	sdkp->capacity = 0;
639
640	return ret;
641}