xref: /openbmc/linux/drivers/scsi/sd_zbc.c (revision f9834f18)
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 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  */
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  */
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  */
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 
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 	sector_t capacity = logical_to_sectors(sdkp->device, sdkp->capacity);
165 	unsigned int nr, i;
166 	unsigned char *buf;
167 	size_t offset, buflen = 0;
168 	int zone_idx = 0;
169 	int ret;
170 
171 	if (!sd_is_zoned(sdkp))
172 		/* Not a zoned device */
173 		return -EOPNOTSUPP;
174 
175 	if (!capacity)
176 		/* Device gone or invalid */
177 		return -ENODEV;
178 
179 	buf = sd_zbc_alloc_report_buffer(sdkp, nr_zones, &buflen);
180 	if (!buf)
181 		return -ENOMEM;
182 
183 	while (zone_idx < nr_zones && sector < capacity) {
184 		ret = sd_zbc_do_report_zones(sdkp, buf, buflen,
185 				sectors_to_logical(sdkp->device, sector), true);
186 		if (ret)
187 			goto out;
188 
189 		offset = 0;
190 		nr = min(nr_zones, get_unaligned_be32(&buf[0]) / 64);
191 		if (!nr)
192 			break;
193 
194 		for (i = 0; i < nr && zone_idx < nr_zones; i++) {
195 			offset += 64;
196 			ret = sd_zbc_parse_report(sdkp, buf + offset, zone_idx,
197 						  cb, data);
198 			if (ret)
199 				goto out;
200 			zone_idx++;
201 		}
202 
203 		sector += sd_zbc_zone_sectors(sdkp) * i;
204 	}
205 
206 	ret = zone_idx;
207 out:
208 	kvfree(buf);
209 	return ret;
210 }
211 
212 /**
213  * sd_zbc_setup_zone_mgmt_cmnd - Prepare a zone ZBC_OUT command. The operations
214  *			can be RESET WRITE POINTER, OPEN, CLOSE or FINISH.
215  * @cmd: the command to setup
216  * @op: Operation to be performed
217  * @all: All zones control
218  *
219  * Called from sd_init_command() for REQ_OP_ZONE_RESET, REQ_OP_ZONE_RESET_ALL,
220  * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE or REQ_OP_ZONE_FINISH requests.
221  */
222 blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
223 					 unsigned char op, bool all)
224 {
225 	struct request *rq = cmd->request;
226 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
227 	sector_t sector = blk_rq_pos(rq);
228 	sector_t block = sectors_to_logical(sdkp->device, sector);
229 
230 	if (!sd_is_zoned(sdkp))
231 		/* Not a zoned device */
232 		return BLK_STS_IOERR;
233 
234 	if (sdkp->device->changed)
235 		return BLK_STS_IOERR;
236 
237 	if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
238 		/* Unaligned request */
239 		return BLK_STS_IOERR;
240 
241 	cmd->cmd_len = 16;
242 	memset(cmd->cmnd, 0, cmd->cmd_len);
243 	cmd->cmnd[0] = ZBC_OUT;
244 	cmd->cmnd[1] = op;
245 	if (all)
246 		cmd->cmnd[14] = 0x1;
247 	else
248 		put_unaligned_be64(block, &cmd->cmnd[2]);
249 
250 	rq->timeout = SD_TIMEOUT;
251 	cmd->sc_data_direction = DMA_NONE;
252 	cmd->transfersize = 0;
253 	cmd->allowed = 0;
254 
255 	return BLK_STS_OK;
256 }
257 
258 /**
259  * sd_zbc_complete - ZBC command post processing.
260  * @cmd: Completed command
261  * @good_bytes: Command reply bytes
262  * @sshdr: command sense header
263  *
264  * Called from sd_done(). Process report zones reply and handle reset zone
265  * and write commands errors.
266  */
267 void sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
268 		     struct scsi_sense_hdr *sshdr)
269 {
270 	int result = cmd->result;
271 	struct request *rq = cmd->request;
272 
273 	if (op_is_zone_mgmt(req_op(rq)) &&
274 	    result &&
275 	    sshdr->sense_key == ILLEGAL_REQUEST &&
276 	    sshdr->asc == 0x24) {
277 		/*
278 		 * INVALID FIELD IN CDB error: a zone management command was
279 		 * attempted on a conventional zone. Nothing to worry about,
280 		 * so be quiet about the error.
281 		 */
282 		rq->rq_flags |= RQF_QUIET;
283 	}
284 }
285 
286 /**
287  * sd_zbc_check_zoned_characteristics - Check zoned block device characteristics
288  * @sdkp: Target disk
289  * @buf: Buffer where to store the VPD page data
290  *
291  * Read VPD page B6, get information and check that reads are unconstrained.
292  */
293 static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
294 					      unsigned char *buf)
295 {
296 
297 	if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
298 		sd_printk(KERN_NOTICE, sdkp,
299 			  "Read zoned characteristics VPD page failed\n");
300 		return -ENODEV;
301 	}
302 
303 	if (sdkp->device->type != TYPE_ZBC) {
304 		/* Host-aware */
305 		sdkp->urswrz = 1;
306 		sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]);
307 		sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]);
308 		sdkp->zones_max_open = 0;
309 	} else {
310 		/* Host-managed */
311 		sdkp->urswrz = buf[4] & 1;
312 		sdkp->zones_optimal_open = 0;
313 		sdkp->zones_optimal_nonseq = 0;
314 		sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
315 	}
316 
317 	/*
318 	 * Check for unconstrained reads: host-managed devices with
319 	 * constrained reads (drives failing read after write pointer)
320 	 * are not supported.
321 	 */
322 	if (!sdkp->urswrz) {
323 		if (sdkp->first_scan)
324 			sd_printk(KERN_NOTICE, sdkp,
325 			  "constrained reads devices are not supported\n");
326 		return -ENODEV;
327 	}
328 
329 	return 0;
330 }
331 
332 /**
333  * sd_zbc_check_capacity - Check the device capacity
334  * @sdkp: Target disk
335  * @buf: command buffer
336  * @zblock: zone size in number of blocks
337  *
338  * Get the device zone size and check that the device capacity as reported
339  * by READ CAPACITY matches the max_lba value (plus one) of the report zones
340  * command reply for devices with RC_BASIS == 0.
341  *
342  * Returns 0 upon success or an error code upon failure.
343  */
344 static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf,
345 				 u32 *zblocks)
346 {
347 	u64 zone_blocks;
348 	sector_t max_lba;
349 	unsigned char *rec;
350 	int ret;
351 
352 	/* Do a report zone to get max_lba and the size of the first zone */
353 	ret = sd_zbc_do_report_zones(sdkp, buf, SD_BUF_SIZE, 0, false);
354 	if (ret)
355 		return ret;
356 
357 	if (sdkp->rc_basis == 0) {
358 		/* The max_lba field is the capacity of this device */
359 		max_lba = get_unaligned_be64(&buf[8]);
360 		if (sdkp->capacity != max_lba + 1) {
361 			if (sdkp->first_scan)
362 				sd_printk(KERN_WARNING, sdkp,
363 					"Changing capacity from %llu to max LBA+1 %llu\n",
364 					(unsigned long long)sdkp->capacity,
365 					(unsigned long long)max_lba + 1);
366 			sdkp->capacity = max_lba + 1;
367 		}
368 	}
369 
370 	/* Get the size of the first reported zone */
371 	rec = buf + 64;
372 	zone_blocks = get_unaligned_be64(&rec[8]);
373 	if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
374 		if (sdkp->first_scan)
375 			sd_printk(KERN_NOTICE, sdkp,
376 				  "Zone size too large\n");
377 		return -EFBIG;
378 	}
379 
380 	*zblocks = zone_blocks;
381 
382 	return 0;
383 }
384 
385 int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
386 {
387 	struct gendisk *disk = sdkp->disk;
388 	unsigned int nr_zones;
389 	u32 zone_blocks = 0;
390 	int ret;
391 
392 	if (!sd_is_zoned(sdkp))
393 		/*
394 		 * Device managed or normal SCSI disk,
395 		 * no special handling required
396 		 */
397 		return 0;
398 
399 	/* Check zoned block device characteristics (unconstrained reads) */
400 	ret = sd_zbc_check_zoned_characteristics(sdkp, buf);
401 	if (ret)
402 		goto err;
403 
404 	/* Check the device capacity reported by report zones */
405 	ret = sd_zbc_check_capacity(sdkp, buf, &zone_blocks);
406 	if (ret != 0)
407 		goto err;
408 
409 	/* The drive satisfies the kernel restrictions: set it up */
410 	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, sdkp->disk->queue);
411 	blk_queue_required_elevator_features(sdkp->disk->queue,
412 					     ELEVATOR_F_ZBD_SEQ_WRITE);
413 	nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
414 
415 	/* READ16/WRITE16 is mandatory for ZBC disks */
416 	sdkp->device->use_16_for_rw = 1;
417 	sdkp->device->use_10_for_rw = 0;
418 
419 	/*
420 	 * Revalidate the disk zone bitmaps once the block device capacity is
421 	 * set on the second revalidate execution during disk scan and if
422 	 * something changed when executing a normal revalidate.
423 	 */
424 	if (sdkp->first_scan) {
425 		sdkp->zone_blocks = zone_blocks;
426 		sdkp->nr_zones = nr_zones;
427 		return 0;
428 	}
429 
430 	if (sdkp->zone_blocks != zone_blocks ||
431 	    sdkp->nr_zones != nr_zones ||
432 	    disk->queue->nr_zones != nr_zones) {
433 		ret = blk_revalidate_disk_zones(disk);
434 		if (ret != 0)
435 			goto err;
436 		sdkp->zone_blocks = zone_blocks;
437 		sdkp->nr_zones = nr_zones;
438 	}
439 
440 	return 0;
441 
442 err:
443 	sdkp->capacity = 0;
444 
445 	return ret;
446 }
447 
448 void sd_zbc_print_zones(struct scsi_disk *sdkp)
449 {
450 	if (!sd_is_zoned(sdkp) || !sdkp->capacity)
451 		return;
452 
453 	if (sdkp->capacity & (sdkp->zone_blocks - 1))
454 		sd_printk(KERN_NOTICE, sdkp,
455 			  "%u zones of %u logical blocks + 1 runt zone\n",
456 			  sdkp->nr_zones - 1,
457 			  sdkp->zone_blocks);
458 	else
459 		sd_printk(KERN_NOTICE, sdkp,
460 			  "%u zones of %u logical blocks\n",
461 			  sdkp->nr_zones,
462 			  sdkp->zone_blocks);
463 }
464