• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
sd_zbc_parse_report(struct scsi_disk * sdkp,u8 * buf,unsigned int idx,report_zones_cb cb,void * data)22 static int sd_zbc_parse_report(struct scsi_disk *sdkp, u8 *buf,
23 			       unsigned int idx, report_zones_cb cb, void *data)
24 {
25 	struct scsi_device *sdp = sdkp->device;
26 	struct blk_zone zone = { 0 };
27 
28 	zone.type = buf[0] & 0x0f;
29 	zone.cond = (buf[1] >> 4) & 0xf;
30 	if (buf[1] & 0x01)
31 		zone.reset = 1;
32 	if (buf[1] & 0x02)
33 		zone.non_seq = 1;
34 
35 	zone.len = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
36 	zone.start = logical_to_sectors(sdp, get_unaligned_be64(&buf[16]));
37 	zone.wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24]));
38 	if (zone.type != ZBC_ZONE_TYPE_CONV &&
39 	    zone.cond == ZBC_ZONE_COND_FULL)
40 		zone.wp = zone.start + zone.len;
41 
42 	return cb(&zone, idx, data);
43 }
44 
45 /**
46  * sd_zbc_do_report_zones - Issue a REPORT ZONES scsi command.
47  * @sdkp: The target disk
48  * @buf: vmalloc-ed buffer to use for the reply
49  * @buflen: the buffer size
50  * @lba: Start LBA of the report
51  * @partial: Do partial report
52  *
53  * For internal use during device validation.
54  * Using partial=true can significantly speed up execution of a report zones
55  * command because the disk does not have to count all possible report matching
56  * zones and will only report the count of zones fitting in the command reply
57  * buffer.
58  */
sd_zbc_do_report_zones(struct scsi_disk * sdkp,unsigned char * buf,unsigned int buflen,sector_t lba,bool partial)59 static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
60 				  unsigned int buflen, sector_t lba,
61 				  bool partial)
62 {
63 	struct scsi_device *sdp = sdkp->device;
64 	const int timeout = sdp->request_queue->rq_timeout;
65 	struct scsi_sense_hdr sshdr;
66 	unsigned char cmd[16];
67 	unsigned int rep_len;
68 	int result;
69 
70 	memset(cmd, 0, 16);
71 	cmd[0] = ZBC_IN;
72 	cmd[1] = ZI_REPORT_ZONES;
73 	put_unaligned_be64(lba, &cmd[2]);
74 	put_unaligned_be32(buflen, &cmd[10]);
75 	if (partial)
76 		cmd[14] = ZBC_REPORT_ZONE_PARTIAL;
77 
78 	result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
79 				  buf, buflen, &sshdr,
80 				  timeout, SD_MAX_RETRIES, NULL);
81 	if (result) {
82 		sd_printk(KERN_ERR, sdkp,
83 			  "REPORT ZONES start lba %llu failed\n", lba);
84 		sd_print_result(sdkp, "REPORT ZONES", result);
85 		if (driver_byte(result) == DRIVER_SENSE &&
86 		    scsi_sense_valid(&sshdr))
87 			sd_print_sense_hdr(sdkp, &sshdr);
88 		return -EIO;
89 	}
90 
91 	rep_len = get_unaligned_be32(&buf[0]);
92 	if (rep_len < 64) {
93 		sd_printk(KERN_ERR, sdkp,
94 			  "REPORT ZONES report invalid length %u\n",
95 			  rep_len);
96 		return -EIO;
97 	}
98 
99 	return 0;
100 }
101 
102 /**
103  * Allocate a buffer for report zones reply.
104  * @sdkp: The target disk
105  * @nr_zones: Maximum number of zones to report
106  * @buflen: Size of the buffer allocated
107  *
108  * Try to allocate a reply buffer for the number of requested zones.
109  * The size of the buffer allocated may be smaller than requested to
110  * satify the device constraint (max_hw_sectors, max_segments, etc).
111  *
112  * Return the address of the allocated buffer and update @buflen with
113  * the size of the allocated buffer.
114  */
sd_zbc_alloc_report_buffer(struct scsi_disk * sdkp,unsigned int nr_zones,size_t * buflen)115 static void *sd_zbc_alloc_report_buffer(struct scsi_disk *sdkp,
116 					unsigned int nr_zones, size_t *buflen)
117 {
118 	struct request_queue *q = sdkp->disk->queue;
119 	size_t bufsize;
120 	void *buf;
121 
122 	/*
123 	 * Report zone buffer size should be at most 64B times the number of
124 	 * zones requested plus the 64B reply header, but should be at least
125 	 * SECTOR_SIZE for ATA devices.
126 	 * Make sure that this size does not exceed the hardware capabilities.
127 	 * Furthermore, since the report zone command cannot be split, make
128 	 * sure that the allocated buffer can always be mapped by limiting the
129 	 * number of pages allocated to the HBA max segments limit.
130 	 */
131 	nr_zones = min(nr_zones, sdkp->nr_zones);
132 	bufsize = roundup((nr_zones + 1) * 64, SECTOR_SIZE);
133 	bufsize = min_t(size_t, bufsize,
134 			queue_max_hw_sectors(q) << SECTOR_SHIFT);
135 	bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
136 
137 	while (bufsize >= SECTOR_SIZE) {
138 		buf = __vmalloc(bufsize,
139 				GFP_KERNEL | __GFP_ZERO | __GFP_NORETRY,
140 				PAGE_KERNEL);
141 		if (buf) {
142 			*buflen = bufsize;
143 			return buf;
144 		}
145 		bufsize >>= 1;
146 	}
147 
148 	return NULL;
149 }
150 
151 /**
152  * sd_zbc_zone_sectors - Get the device zone size in number of 512B sectors.
153  * @sdkp: The target disk
154  */
sd_zbc_zone_sectors(struct scsi_disk * sdkp)155 static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
156 {
157 	return logical_to_sectors(sdkp->device, sdkp->zone_blocks);
158 }
159 
sd_zbc_report_zones(struct gendisk * disk,sector_t sector,unsigned int nr_zones,report_zones_cb cb,void * data)160 int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
161 			unsigned int nr_zones, report_zones_cb cb, void *data)
162 {
163 	struct scsi_disk *sdkp = scsi_disk(disk);
164 	unsigned int nr, i;
165 	unsigned char *buf;
166 	size_t offset, buflen = 0;
167 	int zone_idx = 0;
168 	int ret;
169 
170 	if (!sd_is_zoned(sdkp))
171 		/* Not a zoned device */
172 		return -EOPNOTSUPP;
173 
174 	buf = sd_zbc_alloc_report_buffer(sdkp, nr_zones, &buflen);
175 	if (!buf)
176 		return -ENOMEM;
177 
178 	while (zone_idx < nr_zones && sector < get_capacity(disk)) {
179 		ret = sd_zbc_do_report_zones(sdkp, buf, buflen,
180 				sectors_to_logical(sdkp->device, sector), true);
181 		if (ret)
182 			goto out;
183 
184 		offset = 0;
185 		nr = min(nr_zones, get_unaligned_be32(&buf[0]) / 64);
186 		if (!nr)
187 			break;
188 
189 		for (i = 0; i < nr && zone_idx < nr_zones; i++) {
190 			offset += 64;
191 			ret = sd_zbc_parse_report(sdkp, buf + offset, zone_idx,
192 						  cb, data);
193 			if (ret)
194 				goto out;
195 			zone_idx++;
196 		}
197 
198 		sector += sd_zbc_zone_sectors(sdkp) * i;
199 	}
200 
201 	ret = zone_idx;
202 out:
203 	kvfree(buf);
204 	return ret;
205 }
206 
207 /**
208  * sd_zbc_setup_zone_mgmt_cmnd - Prepare a zone ZBC_OUT command. The operations
209  *			can be RESET WRITE POINTER, OPEN, CLOSE or FINISH.
210  * @cmd: the command to setup
211  * @op: Operation to be performed
212  * @all: All zones control
213  *
214  * Called from sd_init_command() for REQ_OP_ZONE_RESET, REQ_OP_ZONE_RESET_ALL,
215  * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE or REQ_OP_ZONE_FINISH requests.
216  */
sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd * cmd,unsigned char op,bool all)217 blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
218 					 unsigned char op, bool all)
219 {
220 	struct request *rq = cmd->request;
221 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
222 	sector_t sector = blk_rq_pos(rq);
223 	sector_t block = sectors_to_logical(sdkp->device, sector);
224 
225 	if (!sd_is_zoned(sdkp))
226 		/* Not a zoned device */
227 		return BLK_STS_IOERR;
228 
229 	if (sdkp->device->changed)
230 		return BLK_STS_IOERR;
231 
232 	if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
233 		/* Unaligned request */
234 		return BLK_STS_IOERR;
235 
236 	cmd->cmd_len = 16;
237 	memset(cmd->cmnd, 0, cmd->cmd_len);
238 	cmd->cmnd[0] = ZBC_OUT;
239 	cmd->cmnd[1] = op;
240 	if (all)
241 		cmd->cmnd[14] = 0x1;
242 	else
243 		put_unaligned_be64(block, &cmd->cmnd[2]);
244 
245 	rq->timeout = SD_TIMEOUT;
246 	cmd->sc_data_direction = DMA_NONE;
247 	cmd->transfersize = 0;
248 	cmd->allowed = 0;
249 
250 	return BLK_STS_OK;
251 }
252 
253 /**
254  * sd_zbc_complete - ZBC command post processing.
255  * @cmd: Completed command
256  * @good_bytes: Command reply bytes
257  * @sshdr: command sense header
258  *
259  * Called from sd_done(). Process report zones reply and handle reset zone
260  * and write commands errors.
261  */
sd_zbc_complete(struct scsi_cmnd * cmd,unsigned int good_bytes,struct scsi_sense_hdr * sshdr)262 void sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
263 		     struct scsi_sense_hdr *sshdr)
264 {
265 	int result = cmd->result;
266 	struct request *rq = cmd->request;
267 
268 	if (op_is_zone_mgmt(req_op(rq)) &&
269 	    result &&
270 	    sshdr->sense_key == ILLEGAL_REQUEST &&
271 	    sshdr->asc == 0x24) {
272 		/*
273 		 * INVALID FIELD IN CDB error: a zone management command was
274 		 * attempted on a conventional zone. Nothing to worry about,
275 		 * so be quiet about the error.
276 		 */
277 		rq->rq_flags |= RQF_QUIET;
278 	}
279 }
280 
281 /**
282  * sd_zbc_check_zoned_characteristics - Check zoned block device characteristics
283  * @sdkp: Target disk
284  * @buf: Buffer where to store the VPD page data
285  *
286  * Read VPD page B6, get information and check that reads are unconstrained.
287  */
sd_zbc_check_zoned_characteristics(struct scsi_disk * sdkp,unsigned char * buf)288 static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
289 					      unsigned char *buf)
290 {
291 
292 	if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
293 		sd_printk(KERN_NOTICE, sdkp,
294 			  "Read zoned characteristics VPD page failed\n");
295 		return -ENODEV;
296 	}
297 
298 	if (sdkp->device->type != TYPE_ZBC) {
299 		/* Host-aware */
300 		sdkp->urswrz = 1;
301 		sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]);
302 		sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]);
303 		sdkp->zones_max_open = 0;
304 	} else {
305 		/* Host-managed */
306 		sdkp->urswrz = buf[4] & 1;
307 		sdkp->zones_optimal_open = 0;
308 		sdkp->zones_optimal_nonseq = 0;
309 		sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
310 	}
311 
312 	/*
313 	 * Check for unconstrained reads: host-managed devices with
314 	 * constrained reads (drives failing read after write pointer)
315 	 * are not supported.
316 	 */
317 	if (!sdkp->urswrz) {
318 		if (sdkp->first_scan)
319 			sd_printk(KERN_NOTICE, sdkp,
320 			  "constrained reads devices are not supported\n");
321 		return -ENODEV;
322 	}
323 
324 	return 0;
325 }
326 
327 /**
328  * sd_zbc_check_zones - Check the device capacity and zone sizes
329  * @sdkp: Target disk
330  *
331  * Check that the device capacity as reported by READ CAPACITY matches the
332  * max_lba value (plus one)of the report zones command reply. Also check that
333  * all zones of the device have an equal size, only allowing the last zone of
334  * the disk to have a smaller size (runt zone). The zone size must also be a
335  * power of two.
336  *
337  * Returns the zone size in number of blocks upon success or an error code
338  * upon failure.
339  */
sd_zbc_check_zones(struct scsi_disk * sdkp,unsigned char * buf,u32 * zblocks)340 static int sd_zbc_check_zones(struct scsi_disk *sdkp, unsigned char *buf,
341 			      u32 *zblocks)
342 {
343 	u64 zone_blocks = 0;
344 	sector_t max_lba;
345 	unsigned char *rec;
346 	int ret;
347 
348 	/* Do a report zone to get max_lba and the size of the first zone */
349 	ret = sd_zbc_do_report_zones(sdkp, buf, SD_BUF_SIZE, 0, false);
350 	if (ret)
351 		return ret;
352 
353 	if (sdkp->rc_basis == 0) {
354 		/* The max_lba field is the capacity of this device */
355 		max_lba = get_unaligned_be64(&buf[8]);
356 		if (sdkp->capacity != max_lba + 1) {
357 			if (sdkp->first_scan)
358 				sd_printk(KERN_WARNING, sdkp,
359 					"Changing capacity from %llu to max LBA+1 %llu\n",
360 					(unsigned long long)sdkp->capacity,
361 					(unsigned long long)max_lba + 1);
362 			sdkp->capacity = max_lba + 1;
363 		}
364 	}
365 
366 	/* Parse REPORT ZONES header */
367 	rec = buf + 64;
368 	zone_blocks = get_unaligned_be64(&rec[8]);
369 	if (!zone_blocks || !is_power_of_2(zone_blocks)) {
370 		if (sdkp->first_scan)
371 			sd_printk(KERN_NOTICE, sdkp,
372 				  "Devices with non power of 2 zone "
373 				  "size are not supported\n");
374 		return -ENODEV;
375 	}
376 
377 	if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
378 		if (sdkp->first_scan)
379 			sd_printk(KERN_NOTICE, sdkp,
380 				  "Zone size too large\n");
381 		return -EFBIG;
382 	}
383 
384 	*zblocks = zone_blocks;
385 
386 	return 0;
387 }
388 
sd_zbc_read_zones(struct scsi_disk * sdkp,unsigned char * buf)389 int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
390 {
391 	struct gendisk *disk = sdkp->disk;
392 	unsigned int nr_zones;
393 	u32 zone_blocks = 0;
394 	int ret;
395 
396 	if (!sd_is_zoned(sdkp))
397 		/*
398 		 * Device managed or normal SCSI disk,
399 		 * no special handling required
400 		 */
401 		return 0;
402 
403 	/* Check zoned block device characteristics (unconstrained reads) */
404 	ret = sd_zbc_check_zoned_characteristics(sdkp, buf);
405 	if (ret)
406 		goto err;
407 
408 	/*
409 	 * Check zone size: only devices with a constant zone size (except
410 	 * an eventual last runt zone) that is a power of 2 are supported.
411 	 */
412 	ret = sd_zbc_check_zones(sdkp, buf, &zone_blocks);
413 	if (ret != 0)
414 		goto err;
415 
416 	/* The drive satisfies the kernel restrictions: set it up */
417 	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, sdkp->disk->queue);
418 	blk_queue_required_elevator_features(sdkp->disk->queue,
419 					     ELEVATOR_F_ZBD_SEQ_WRITE);
420 	nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
421 
422 	/* READ16/WRITE16 is mandatory for ZBC disks */
423 	sdkp->device->use_16_for_rw = 1;
424 	sdkp->device->use_10_for_rw = 0;
425 
426 	/*
427 	 * Revalidate the disk zone bitmaps once the block device capacity is
428 	 * set on the second revalidate execution during disk scan and if
429 	 * something changed when executing a normal revalidate.
430 	 */
431 	if (sdkp->first_scan) {
432 		sdkp->zone_blocks = zone_blocks;
433 		sdkp->nr_zones = nr_zones;
434 		return 0;
435 	}
436 
437 	if (sdkp->zone_blocks != zone_blocks ||
438 	    sdkp->nr_zones != nr_zones ||
439 	    disk->queue->nr_zones != nr_zones) {
440 		ret = blk_revalidate_disk_zones(disk);
441 		if (ret != 0)
442 			goto err;
443 		sdkp->zone_blocks = zone_blocks;
444 		sdkp->nr_zones = nr_zones;
445 	}
446 
447 	return 0;
448 
449 err:
450 	sdkp->capacity = 0;
451 
452 	return ret;
453 }
454 
sd_zbc_print_zones(struct scsi_disk * sdkp)455 void sd_zbc_print_zones(struct scsi_disk *sdkp)
456 {
457 	if (!sd_is_zoned(sdkp) || !sdkp->capacity)
458 		return;
459 
460 	if (sdkp->capacity & (sdkp->zone_blocks - 1))
461 		sd_printk(KERN_NOTICE, sdkp,
462 			  "%u zones of %u logical blocks + 1 runt zone\n",
463 			  sdkp->nr_zones - 1,
464 			  sdkp->zone_blocks);
465 	else
466 		sd_printk(KERN_NOTICE, sdkp,
467 			  "%u zones of %u logical blocks\n",
468 			  sdkp->nr_zones,
469 			  sdkp->zone_blocks);
470 }
471