blob: 250cb76ee61532ebd498cbfd3f89f13f22191501 [file] [log] [blame]
Christoph Hellwig3dcf60bc2019-04-30 14:42:43 -04001// SPDX-License-Identifier: GPL-2.0
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +09002/*
3 * Zoned block device handling
4 *
5 * Copyright (c) 2015, Hannes Reinecke
6 * Copyright (c) 2015, SUSE Linux GmbH
7 *
8 * Copyright (c) 2016, Damien Le Moal
9 * Copyright (c) 2016, Western Digital
10 */
11
12#include <linux/kernel.h>
13#include <linux/module.h>
14#include <linux/rbtree.h>
15#include <linux/blkdev.h>
Damien Le Moalbf505452018-10-12 19:08:50 +090016#include <linux/blk-mq.h>
Damien Le Moal26202922019-07-01 14:09:18 +090017#include <linux/mm.h>
18#include <linux/vmalloc.h>
Damien Le Moalbd976e52019-07-01 14:09:16 +090019#include <linux/sched/mm.h>
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +090020
Damien Le Moala2d6b3a2018-10-12 19:08:47 +090021#include "blk.h"
22
Chaitanya Kulkarni02694e82020-03-25 10:49:54 -070023#define ZONE_COND_NAME(name) [BLK_ZONE_COND_##name] = #name
24static const char *const zone_cond_name[] = {
25 ZONE_COND_NAME(NOT_WP),
26 ZONE_COND_NAME(EMPTY),
27 ZONE_COND_NAME(IMP_OPEN),
28 ZONE_COND_NAME(EXP_OPEN),
29 ZONE_COND_NAME(CLOSED),
30 ZONE_COND_NAME(READONLY),
31 ZONE_COND_NAME(FULL),
32 ZONE_COND_NAME(OFFLINE),
33};
34#undef ZONE_COND_NAME
35
36/**
37 * blk_zone_cond_str - Return string XXX in BLK_ZONE_COND_XXX.
38 * @zone_cond: BLK_ZONE_COND_XXX.
39 *
40 * Description: Centralize block layer function to convert BLK_ZONE_COND_XXX
41 * into string format. Useful in the debugging and tracing zone conditions. For
42 * invalid BLK_ZONE_COND_XXX it returns string "UNKNOWN".
43 */
44const char *blk_zone_cond_str(enum blk_zone_cond zone_cond)
45{
46 static const char *zone_cond_str = "UNKNOWN";
47
48 if (zone_cond < ARRAY_SIZE(zone_cond_name) && zone_cond_name[zone_cond])
49 zone_cond_str = zone_cond_name[zone_cond];
50
51 return zone_cond_str;
52}
53EXPORT_SYMBOL_GPL(blk_zone_cond_str);
54
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +090055/*
Christoph Hellwig6cc77e92017-12-21 15:43:38 +090056 * Return true if a request is a write requests that needs zone write locking.
57 */
58bool blk_req_needs_zone_write_lock(struct request *rq)
59{
60 if (!rq->q->seq_zones_wlock)
61 return false;
62
63 if (blk_rq_is_passthrough(rq))
64 return false;
65
66 switch (req_op(rq)) {
67 case REQ_OP_WRITE_ZEROES:
68 case REQ_OP_WRITE_SAME:
69 case REQ_OP_WRITE:
70 return blk_rq_zone_is_seq(rq);
71 default:
72 return false;
73 }
74}
75EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
76
Johannes Thumshirn1392d372020-05-12 17:55:48 +090077bool blk_req_zone_write_trylock(struct request *rq)
78{
79 unsigned int zno = blk_rq_zone_no(rq);
80
81 if (test_and_set_bit(zno, rq->q->seq_zones_wlock))
82 return false;
83
84 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
85 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
86
87 return true;
88}
89EXPORT_SYMBOL_GPL(blk_req_zone_write_trylock);
90
Christoph Hellwig6cc77e92017-12-21 15:43:38 +090091void __blk_req_zone_write_lock(struct request *rq)
92{
93 if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
94 rq->q->seq_zones_wlock)))
95 return;
96
97 WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
98 rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
99}
100EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
101
102void __blk_req_zone_write_unlock(struct request *rq)
103{
104 rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED;
105 if (rq->q->seq_zones_wlock)
106 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq),
107 rq->q->seq_zones_wlock));
108}
109EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
110
Damien Le Moala91e1382018-10-12 19:08:43 +0900111/**
112 * blkdev_nr_zones - Get number of zones
Christoph Hellwig9b38bb42019-12-03 10:39:04 +0100113 * @disk: Target gendisk
Damien Le Moala91e1382018-10-12 19:08:43 +0900114 *
Christoph Hellwig9b38bb42019-12-03 10:39:04 +0100115 * Return the total number of zones of a zoned block device. For a block
116 * device without zone capabilities, the number of zones is always 0.
Damien Le Moala91e1382018-10-12 19:08:43 +0900117 */
Christoph Hellwig9b38bb42019-12-03 10:39:04 +0100118unsigned int blkdev_nr_zones(struct gendisk *disk)
Damien Le Moala91e1382018-10-12 19:08:43 +0900119{
Christoph Hellwig9b38bb42019-12-03 10:39:04 +0100120 sector_t zone_sectors = blk_queue_zone_sectors(disk->queue);
Damien Le Moala91e1382018-10-12 19:08:43 +0900121
Christoph Hellwig9b38bb42019-12-03 10:39:04 +0100122 if (!blk_queue_is_zoned(disk->queue))
Damien Le Moala91e1382018-10-12 19:08:43 +0900123 return 0;
Christoph Hellwig9b38bb42019-12-03 10:39:04 +0100124 return (get_capacity(disk) + zone_sectors - 1) >> ilog2(zone_sectors);
Damien Le Moala91e1382018-10-12 19:08:43 +0900125}
126EXPORT_SYMBOL_GPL(blkdev_nr_zones);
127
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900128/**
129 * blkdev_report_zones - Get zones information
130 * @bdev: Target block device
131 * @sector: Sector from which to report zones
Christoph Hellwigd4100352019-11-11 11:39:30 +0900132 * @nr_zones: Maximum number of zones to report
133 * @cb: Callback function called for each reported zone
134 * @data: Private data for the callback
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900135 *
136 * Description:
Christoph Hellwigd4100352019-11-11 11:39:30 +0900137 * Get zone information starting from the zone containing @sector for at most
138 * @nr_zones, and call @cb for each zone reported by the device.
139 * To report all zones in a device starting from @sector, the BLK_ALL_ZONES
140 * constant can be passed to @nr_zones.
141 * Returns the number of zones reported by the device, or a negative errno
142 * value in case of failure.
143 *
144 * Note: The caller must use memalloc_noXX_save/restore() calls to control
145 * memory allocations done within this function.
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900146 */
Christoph Hellwige76239a2018-10-12 19:08:49 +0900147int blkdev_report_zones(struct block_device *bdev, sector_t sector,
Christoph Hellwigd4100352019-11-11 11:39:30 +0900148 unsigned int nr_zones, report_zones_cb cb, void *data)
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900149{
Damien Le Moalceeb3732019-11-11 11:39:24 +0900150 struct gendisk *disk = bdev->bd_disk;
Damien Le Moal5eac3eb2019-11-11 11:39:25 +0900151 sector_t capacity = get_capacity(disk);
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900152
Christoph Hellwigd4100352019-11-11 11:39:30 +0900153 if (!blk_queue_is_zoned(bdev_get_queue(bdev)) ||
154 WARN_ON_ONCE(!disk->fops->report_zones))
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900155 return -EOPNOTSUPP;
156
Christoph Hellwigd4100352019-11-11 11:39:30 +0900157 if (!nr_zones || sector >= capacity)
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900158 return 0;
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900159
Christoph Hellwigd4100352019-11-11 11:39:30 +0900160 return disk->fops->report_zones(disk, sector, nr_zones, cb, data);
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900161}
162EXPORT_SYMBOL_GPL(blkdev_report_zones);
163
Chaitanya Kulkarni6e33dbf2019-08-01 10:26:36 -0700164static inline bool blkdev_allow_reset_all_zones(struct block_device *bdev,
Damien Le Moalc7a1d922019-10-27 23:05:43 +0900165 sector_t sector,
Chaitanya Kulkarni6e33dbf2019-08-01 10:26:36 -0700166 sector_t nr_sectors)
167{
168 if (!blk_queue_zone_resetall(bdev_get_queue(bdev)))
169 return false;
170
Chaitanya Kulkarni6e33dbf2019-08-01 10:26:36 -0700171 /*
Damien Le Moal5eac3eb2019-11-11 11:39:25 +0900172 * REQ_OP_ZONE_RESET_ALL can be executed only if the number of sectors
173 * of the applicable zone range is the entire disk.
Chaitanya Kulkarni6e33dbf2019-08-01 10:26:36 -0700174 */
Damien Le Moal5eac3eb2019-11-11 11:39:25 +0900175 return !sector && nr_sectors == get_capacity(bdev->bd_disk);
Chaitanya Kulkarni6e33dbf2019-08-01 10:26:36 -0700176}
177
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900178/**
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900179 * blkdev_zone_mgmt - Execute a zone management operation on a range of zones
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900180 * @bdev: Target block device
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900181 * @op: Operation to be performed on the zones
182 * @sector: Start sector of the first zone to operate on
183 * @nr_sectors: Number of sectors, should be at least the length of one zone and
184 * must be zone size aligned.
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900185 * @gfp_mask: Memory allocation flags (for bio_alloc)
186 *
187 * Description:
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900188 * Perform the specified operation on the range of zones specified by
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900189 * @sector..@sector+@nr_sectors. Specifying the entire disk sector range
190 * is valid, but the specified range should not contain conventional zones.
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900191 * The operation to execute on each zone can be a zone reset, open, close
192 * or finish request.
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900193 */
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900194int blkdev_zone_mgmt(struct block_device *bdev, enum req_opf op,
195 sector_t sector, sector_t nr_sectors,
196 gfp_t gfp_mask)
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900197{
198 struct request_queue *q = bdev_get_queue(bdev);
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900199 sector_t zone_sectors = blk_queue_zone_sectors(q);
Damien Le Moal5eac3eb2019-11-11 11:39:25 +0900200 sector_t capacity = get_capacity(bdev->bd_disk);
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900201 sector_t end_sector = sector + nr_sectors;
Damien Le Moala2d6b3a2018-10-12 19:08:47 +0900202 struct bio *bio = NULL;
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900203 int ret;
204
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900205 if (!blk_queue_is_zoned(q))
206 return -EOPNOTSUPP;
207
Damien Le Moala2d6b3a2018-10-12 19:08:47 +0900208 if (bdev_read_only(bdev))
209 return -EPERM;
210
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900211 if (!op_is_zone_mgmt(op))
212 return -EOPNOTSUPP;
213
Alexey Dobriyan11bde982020-02-12 20:40:27 +0300214 if (end_sector <= sector || end_sector > capacity)
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900215 /* Out of range */
216 return -EINVAL;
217
218 /* Check alignment (handle eventual smaller last zone) */
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900219 if (sector & (zone_sectors - 1))
220 return -EINVAL;
221
Damien Le Moal5eac3eb2019-11-11 11:39:25 +0900222 if ((nr_sectors & (zone_sectors - 1)) && end_sector != capacity)
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900223 return -EINVAL;
224
225 while (sector < end_sector) {
Damien Le Moala2d6b3a2018-10-12 19:08:47 +0900226 bio = blk_next_bio(bio, 0, gfp_mask);
Christoph Hellwig74d46992017-08-23 19:10:32 +0200227 bio_set_dev(bio, bdev);
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900228
Damien Le Moalc7a1d922019-10-27 23:05:43 +0900229 /*
230 * Special case for the zone reset operation that reset all
231 * zones, this is useful for applications like mkfs.
232 */
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900233 if (op == REQ_OP_ZONE_RESET &&
234 blkdev_allow_reset_all_zones(bdev, sector, nr_sectors)) {
Damien Le Moalfaa44c62021-03-10 18:09:19 +0900235 bio->bi_opf = REQ_OP_ZONE_RESET_ALL | REQ_SYNC;
Damien Le Moalc7a1d922019-10-27 23:05:43 +0900236 break;
237 }
238
Chaitanya Kulkarni8e42d232020-01-07 13:58:17 -0800239 bio->bi_opf = op | REQ_SYNC;
Damien Le Moalc7a1d922019-10-27 23:05:43 +0900240 bio->bi_iter.bi_sector = sector;
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900241 sector += zone_sectors;
242
243 /* This may take a while, so be nice to others */
244 cond_resched();
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900245 }
246
Damien Le Moala2d6b3a2018-10-12 19:08:47 +0900247 ret = submit_bio_wait(bio);
248 bio_put(bio);
249
Damien Le Moala2d6b3a2018-10-12 19:08:47 +0900250 return ret;
Hannes Reinecke6a0cb1b2016-10-18 15:40:33 +0900251}
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900252EXPORT_SYMBOL_GPL(blkdev_zone_mgmt);
Shaun Tancheff3ed05a92016-10-18 15:40:35 +0900253
Christoph Hellwigd4100352019-11-11 11:39:30 +0900254struct zone_report_args {
255 struct blk_zone __user *zones;
256};
257
258static int blkdev_copy_zone_to_user(struct blk_zone *zone, unsigned int idx,
259 void *data)
260{
261 struct zone_report_args *args = data;
262
263 if (copy_to_user(&args->zones[idx], zone, sizeof(struct blk_zone)))
264 return -EFAULT;
265 return 0;
266}
267
Bart Van Assche56c4bdd2018-03-08 15:28:50 -0800268/*
Shaun Tancheff3ed05a92016-10-18 15:40:35 +0900269 * BLKREPORTZONE ioctl processing.
270 * Called from blkdev_ioctl.
271 */
272int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
273 unsigned int cmd, unsigned long arg)
274{
275 void __user *argp = (void __user *)arg;
Christoph Hellwigd4100352019-11-11 11:39:30 +0900276 struct zone_report_args args;
Shaun Tancheff3ed05a92016-10-18 15:40:35 +0900277 struct request_queue *q;
278 struct blk_zone_report rep;
Shaun Tancheff3ed05a92016-10-18 15:40:35 +0900279 int ret;
280
281 if (!argp)
282 return -EINVAL;
283
284 q = bdev_get_queue(bdev);
285 if (!q)
286 return -ENXIO;
287
288 if (!blk_queue_is_zoned(q))
289 return -ENOTTY;
290
291 if (!capable(CAP_SYS_ADMIN))
292 return -EACCES;
293
294 if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report)))
295 return -EFAULT;
296
297 if (!rep.nr_zones)
298 return -EINVAL;
299
Christoph Hellwigd4100352019-11-11 11:39:30 +0900300 args.zones = argp + sizeof(struct blk_zone_report);
301 ret = blkdev_report_zones(bdev, rep.sector, rep.nr_zones,
302 blkdev_copy_zone_to_user, &args);
303 if (ret < 0)
304 return ret;
Bart Van Assche327ea4a2018-05-22 08:27:22 -0700305
Christoph Hellwigd4100352019-11-11 11:39:30 +0900306 rep.nr_zones = ret;
Matias Bjørling82394db2020-06-29 12:06:37 -0700307 rep.flags = BLK_ZONE_REP_CAPACITY;
Christoph Hellwigd4100352019-11-11 11:39:30 +0900308 if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report)))
309 return -EFAULT;
310 return 0;
Shaun Tancheff3ed05a92016-10-18 15:40:35 +0900311}
312
Shin'ichiro Kawasakie5113502021-03-11 16:25:46 +0900313static int blkdev_truncate_zone_range(struct block_device *bdev, fmode_t mode,
314 const struct blk_zone_range *zrange)
315{
316 loff_t start, end;
317
318 if (zrange->sector + zrange->nr_sectors <= zrange->sector ||
319 zrange->sector + zrange->nr_sectors > get_capacity(bdev->bd_disk))
320 /* Out of range */
321 return -EINVAL;
322
323 start = zrange->sector << SECTOR_SHIFT;
324 end = ((zrange->sector + zrange->nr_sectors) << SECTOR_SHIFT) - 1;
325
326 return truncate_bdev_range(bdev, mode, start, end);
327}
328
Bart Van Assche56c4bdd2018-03-08 15:28:50 -0800329/*
Ajay Joshie876df12019-10-27 23:05:46 +0900330 * BLKRESETZONE, BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE ioctl processing.
Shaun Tancheff3ed05a92016-10-18 15:40:35 +0900331 * Called from blkdev_ioctl.
332 */
Ajay Joshie876df12019-10-27 23:05:46 +0900333int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
334 unsigned int cmd, unsigned long arg)
Shaun Tancheff3ed05a92016-10-18 15:40:35 +0900335{
336 void __user *argp = (void __user *)arg;
337 struct request_queue *q;
338 struct blk_zone_range zrange;
Ajay Joshie876df12019-10-27 23:05:46 +0900339 enum req_opf op;
Shin'ichiro Kawasakie5113502021-03-11 16:25:46 +0900340 int ret;
Shaun Tancheff3ed05a92016-10-18 15:40:35 +0900341
342 if (!argp)
343 return -EINVAL;
344
345 q = bdev_get_queue(bdev);
346 if (!q)
347 return -ENXIO;
348
349 if (!blk_queue_is_zoned(q))
350 return -ENOTTY;
351
352 if (!capable(CAP_SYS_ADMIN))
353 return -EACCES;
354
355 if (!(mode & FMODE_WRITE))
356 return -EBADF;
357
358 if (copy_from_user(&zrange, argp, sizeof(struct blk_zone_range)))
359 return -EFAULT;
360
Ajay Joshie876df12019-10-27 23:05:46 +0900361 switch (cmd) {
362 case BLKRESETZONE:
363 op = REQ_OP_ZONE_RESET;
Shin'ichiro Kawasakie5113502021-03-11 16:25:46 +0900364
365 /* Invalidate the page cache, including dirty pages. */
366 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
367 if (ret)
368 return ret;
Ajay Joshie876df12019-10-27 23:05:46 +0900369 break;
370 case BLKOPENZONE:
371 op = REQ_OP_ZONE_OPEN;
372 break;
373 case BLKCLOSEZONE:
374 op = REQ_OP_ZONE_CLOSE;
375 break;
376 case BLKFINISHZONE:
377 op = REQ_OP_ZONE_FINISH;
378 break;
379 default:
380 return -ENOTTY;
381 }
382
Shin'ichiro Kawasakie5113502021-03-11 16:25:46 +0900383 ret = blkdev_zone_mgmt(bdev, op, zrange.sector, zrange.nr_sectors,
384 GFP_KERNEL);
385
386 /*
387 * Invalidate the page cache again for zone reset: writes can only be
388 * direct for zoned devices so concurrent writes would not add any page
389 * to the page cache after/during reset. The page cache may be filled
390 * again due to concurrent reads though and dropping the pages for
391 * these is fine.
392 */
393 if (!ret && cmd == BLKRESETZONE)
394 ret = blkdev_truncate_zone_range(bdev, mode, &zrange);
395
396 return ret;
Shaun Tancheff3ed05a92016-10-18 15:40:35 +0900397}
Damien Le Moalbf505452018-10-12 19:08:50 +0900398
399static inline unsigned long *blk_alloc_zone_bitmap(int node,
400 unsigned int nr_zones)
401{
402 return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long),
403 GFP_NOIO, node);
404}
405
Damien Le Moalbf505452018-10-12 19:08:50 +0900406void blk_queue_free_zone_bitmaps(struct request_queue *q)
407{
Christoph Hellwigf216fdd2019-12-03 10:39:05 +0100408 kfree(q->conv_zones_bitmap);
409 q->conv_zones_bitmap = NULL;
Damien Le Moalbf505452018-10-12 19:08:50 +0900410 kfree(q->seq_zones_wlock);
411 q->seq_zones_wlock = NULL;
412}
413
Christoph Hellwigd4100352019-11-11 11:39:30 +0900414struct blk_revalidate_zone_args {
415 struct gendisk *disk;
Christoph Hellwigf216fdd2019-12-03 10:39:05 +0100416 unsigned long *conv_zones_bitmap;
Christoph Hellwigd4100352019-11-11 11:39:30 +0900417 unsigned long *seq_zones_wlock;
Christoph Hellwige94f5812019-12-03 10:39:06 +0100418 unsigned int nr_zones;
Christoph Hellwig6c6b3542019-12-03 10:39:08 +0100419 sector_t zone_sectors;
Christoph Hellwigd4100352019-11-11 11:39:30 +0900420 sector_t sector;
421};
422
Damien Le Moald9dd7302019-11-11 11:39:22 +0900423/*
424 * Helper function to check the validity of zones of a zoned block device.
425 */
Christoph Hellwigd4100352019-11-11 11:39:30 +0900426static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx,
427 void *data)
Damien Le Moald9dd7302019-11-11 11:39:22 +0900428{
Christoph Hellwigd4100352019-11-11 11:39:30 +0900429 struct blk_revalidate_zone_args *args = data;
430 struct gendisk *disk = args->disk;
Damien Le Moald9dd7302019-11-11 11:39:22 +0900431 struct request_queue *q = disk->queue;
Damien Le Moald9dd7302019-11-11 11:39:22 +0900432 sector_t capacity = get_capacity(disk);
433
434 /*
435 * All zones must have the same size, with the exception on an eventual
436 * smaller last zone.
437 */
Christoph Hellwig6c6b3542019-12-03 10:39:08 +0100438 if (zone->start == 0) {
439 if (zone->len == 0 || !is_power_of_2(zone->len)) {
440 pr_warn("%s: Invalid zoned device with non power of two zone size (%llu)\n",
441 disk->disk_name, zone->len);
442 return -ENODEV;
443 }
Damien Le Moald9dd7302019-11-11 11:39:22 +0900444
Christoph Hellwig6c6b3542019-12-03 10:39:08 +0100445 args->zone_sectors = zone->len;
446 args->nr_zones = (capacity + zone->len - 1) >> ilog2(zone->len);
447 } else if (zone->start + args->zone_sectors < capacity) {
448 if (zone->len != args->zone_sectors) {
449 pr_warn("%s: Invalid zoned device with non constant zone size\n",
450 disk->disk_name);
451 return -ENODEV;
452 }
453 } else {
454 if (zone->len > args->zone_sectors) {
455 pr_warn("%s: Invalid zoned device with larger last zone size\n",
456 disk->disk_name);
457 return -ENODEV;
458 }
Damien Le Moald9dd7302019-11-11 11:39:22 +0900459 }
460
461 /* Check for holes in the zone report */
Christoph Hellwigd4100352019-11-11 11:39:30 +0900462 if (zone->start != args->sector) {
Damien Le Moald9dd7302019-11-11 11:39:22 +0900463 pr_warn("%s: Zone gap at sectors %llu..%llu\n",
Christoph Hellwigd4100352019-11-11 11:39:30 +0900464 disk->disk_name, args->sector, zone->start);
465 return -ENODEV;
Damien Le Moald9dd7302019-11-11 11:39:22 +0900466 }
467
468 /* Check zone type */
469 switch (zone->type) {
470 case BLK_ZONE_TYPE_CONVENTIONAL:
Christoph Hellwige94f5812019-12-03 10:39:06 +0100471 if (!args->conv_zones_bitmap) {
472 args->conv_zones_bitmap =
473 blk_alloc_zone_bitmap(q->node, args->nr_zones);
474 if (!args->conv_zones_bitmap)
475 return -ENOMEM;
476 }
477 set_bit(idx, args->conv_zones_bitmap);
478 break;
Damien Le Moald9dd7302019-11-11 11:39:22 +0900479 case BLK_ZONE_TYPE_SEQWRITE_REQ:
480 case BLK_ZONE_TYPE_SEQWRITE_PREF:
Christoph Hellwige94f5812019-12-03 10:39:06 +0100481 if (!args->seq_zones_wlock) {
482 args->seq_zones_wlock =
483 blk_alloc_zone_bitmap(q->node, args->nr_zones);
484 if (!args->seq_zones_wlock)
485 return -ENOMEM;
486 }
Damien Le Moald9dd7302019-11-11 11:39:22 +0900487 break;
488 default:
489 pr_warn("%s: Invalid zone type 0x%x at sectors %llu\n",
490 disk->disk_name, (int)zone->type, zone->start);
Christoph Hellwigd4100352019-11-11 11:39:30 +0900491 return -ENODEV;
Damien Le Moald9dd7302019-11-11 11:39:22 +0900492 }
493
Christoph Hellwigd4100352019-11-11 11:39:30 +0900494 args->sector += zone->len;
495 return 0;
496}
497
Damien Le Moalbf505452018-10-12 19:08:50 +0900498/**
499 * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
500 * @disk: Target disk
Damien Le Moale7326712020-05-12 17:55:49 +0900501 * @update_driver_data: Callback to update driver data on the frozen disk
Damien Le Moalbf505452018-10-12 19:08:50 +0900502 *
503 * Helper function for low-level device drivers to (re) allocate and initialize
504 * a disk request queue zone bitmaps. This functions should normally be called
Christoph Hellwigae589542019-12-03 10:39:07 +0100505 * within the disk ->revalidate method for blk-mq based drivers. For BIO based
506 * drivers only q->nr_zones needs to be updated so that the sysfs exposed value
507 * is correct.
Damien Le Moale7326712020-05-12 17:55:49 +0900508 * If the @update_driver_data callback function is not NULL, the callback is
509 * executed with the device request queue frozen after all zones have been
510 * checked.
Damien Le Moalbf505452018-10-12 19:08:50 +0900511 */
Damien Le Moale7326712020-05-12 17:55:49 +0900512int blk_revalidate_disk_zones(struct gendisk *disk,
513 void (*update_driver_data)(struct gendisk *disk))
Damien Le Moalbf505452018-10-12 19:08:50 +0900514{
515 struct request_queue *q = disk->queue;
Christoph Hellwige94f5812019-12-03 10:39:06 +0100516 struct blk_revalidate_zone_args args = {
517 .disk = disk,
Christoph Hellwige94f5812019-12-03 10:39:06 +0100518 };
Christoph Hellwig6c6b3542019-12-03 10:39:08 +0100519 unsigned int noio_flag;
520 int ret;
Damien Le Moalbf505452018-10-12 19:08:50 +0900521
Christoph Hellwigc98c3d092019-11-11 11:39:23 +0900522 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
523 return -EIO;
Christoph Hellwigae589542019-12-03 10:39:07 +0100524 if (WARN_ON_ONCE(!queue_is_mq(q)))
525 return -EIO;
Damien Le Moalbf505452018-10-12 19:08:50 +0900526
Johannes Thumshirn1a1206d2020-07-30 20:25:17 +0900527 if (!get_capacity(disk))
528 return -EIO;
529
Christoph Hellwige94f5812019-12-03 10:39:06 +0100530 /*
Christoph Hellwig6c6b3542019-12-03 10:39:08 +0100531 * Ensure that all memory allocations in this context are done as if
532 * GFP_NOIO was specified.
Christoph Hellwige94f5812019-12-03 10:39:06 +0100533 */
Christoph Hellwig6c6b3542019-12-03 10:39:08 +0100534 noio_flag = memalloc_noio_save();
535 ret = disk->fops->report_zones(disk, 0, UINT_MAX,
536 blk_revalidate_zone_cb, &args);
Damien Le Moal2afdeb22020-11-11 16:36:06 +0900537 if (!ret) {
538 pr_warn("%s: No zones reported\n", disk->disk_name);
539 ret = -ENODEV;
540 }
Christoph Hellwig6c6b3542019-12-03 10:39:08 +0100541 memalloc_noio_restore(noio_flag);
Damien Le Moalbd976e52019-07-01 14:09:16 +0900542
Damien Le Moalbf505452018-10-12 19:08:50 +0900543 /*
Damien Le Moal2afdeb22020-11-11 16:36:06 +0900544 * If zones where reported, make sure that the entire disk capacity
545 * has been checked.
546 */
547 if (ret > 0 && args.sector != get_capacity(disk)) {
548 pr_warn("%s: Missing zones from sector %llu\n",
549 disk->disk_name, args.sector);
550 ret = -ENODEV;
551 }
552
553 /*
Christoph Hellwig6c6b3542019-12-03 10:39:08 +0100554 * Install the new bitmaps and update nr_zones only once the queue is
555 * stopped and all I/Os are completed (i.e. a scheduler is not
556 * referencing the bitmaps).
Damien Le Moalbf505452018-10-12 19:08:50 +0900557 */
558 blk_mq_freeze_queue(q);
Damien Le Moal2afdeb22020-11-11 16:36:06 +0900559 if (ret > 0) {
Christoph Hellwig6c6b3542019-12-03 10:39:08 +0100560 blk_queue_chunk_sectors(q, args.zone_sectors);
Christoph Hellwige94f5812019-12-03 10:39:06 +0100561 q->nr_zones = args.nr_zones;
Christoph Hellwigd4100352019-11-11 11:39:30 +0900562 swap(q->seq_zones_wlock, args.seq_zones_wlock);
Christoph Hellwigf216fdd2019-12-03 10:39:05 +0100563 swap(q->conv_zones_bitmap, args.conv_zones_bitmap);
Damien Le Moale7326712020-05-12 17:55:49 +0900564 if (update_driver_data)
565 update_driver_data(disk);
Christoph Hellwigd4100352019-11-11 11:39:30 +0900566 ret = 0;
567 } else {
568 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
569 blk_queue_free_zone_bitmaps(q);
570 }
Damien Le Moalbf505452018-10-12 19:08:50 +0900571 blk_mq_unfreeze_queue(q);
572
Christoph Hellwigd4100352019-11-11 11:39:30 +0900573 kfree(args.seq_zones_wlock);
Christoph Hellwigf216fdd2019-12-03 10:39:05 +0100574 kfree(args.conv_zones_bitmap);
Damien Le Moalbf505452018-10-12 19:08:50 +0900575 return ret;
576}
577EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);
Damien Le Moal508aebb2021-01-28 13:47:32 +0900578
579void blk_queue_clear_zone_settings(struct request_queue *q)
580{
581 blk_mq_freeze_queue(q);
582
583 blk_queue_free_zone_bitmaps(q);
584 blk_queue_flag_clear(QUEUE_FLAG_ZONE_RESETALL, q);
585 q->required_elevator_features &= ~ELEVATOR_F_ZBD_SEQ_WRITE;
586 q->nr_zones = 0;
587 q->max_open_zones = 0;
588 q->max_active_zones = 0;
589 q->limits.chunk_sectors = 0;
590 q->limits.zone_write_granularity = 0;
591 q->limits.max_zone_append_sectors = 0;
592
593 blk_mq_unfreeze_queue(q);
594}