block/blk-zoned.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   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>
  16 #include <linux/blk-mq.h>
  17 #include <linux/mm.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/sched/mm.h>
  20
  21 #include "blk.h"
  22
  23 static inline sector_t blk_zone_start(struct request_queue *q,
  24                                       sector_t sector)
  25 {
  26         sector_t zone_mask = blk_queue_zone_sectors(q) - 1;
  27
  28         return sector & ~zone_mask;
  29 }
  30
  31 /*
  32  * Return true if a request is a write requests that needs zone write locking.
  33  */
  34 bool blk_req_needs_zone_write_lock(struct request *rq)
  35 {
  36         if (!rq->q->seq_zones_wlock)
  37                 return false;
  38
  39         if (blk_rq_is_passthrough(rq))
  40                 return false;
  41
  42         switch (req_op(rq)) {
  43         case REQ_OP_WRITE_ZEROES:
  44         case REQ_OP_WRITE_SAME:
  45         case REQ_OP_WRITE:
  46                 return blk_rq_zone_is_seq(rq);
  47         default:
  48                 return false;
  49         }
  50 }
  51 EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
  52
  53 void __blk_req_zone_write_lock(struct request *rq)
  54 {
  55         if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
  56                                           rq->q->seq_zones_wlock)))
  57                 return;
  58
  59         WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
  60         rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
  61 }
  62 EXPORT_SYMBOL_GPL(__blk_req_zone_write_lock);
  63
  64 void __blk_req_zone_write_unlock(struct request *rq)
  65 {
  66         rq->rq_flags &= ~RQF_ZONE_WRITE_LOCKED;
  67         if (rq->q->seq_zones_wlock)
  68                 WARN_ON_ONCE(!test_and_clear_bit(blk_rq_zone_no(rq),
  69                                                  rq->q->seq_zones_wlock));
  70 }
  71 EXPORT_SYMBOL_GPL(__blk_req_zone_write_unlock);
  72
  73 static inline unsigned int __blkdev_nr_zones(struct request_queue *q,
  74                                              sector_t nr_sectors)
  75 {
  76         sector_t zone_sectors = blk_queue_zone_sectors(q);
  77
  78         return (nr_sectors + zone_sectors - 1) >> ilog2(zone_sectors);
  79 }
  80
  81 /**
  82  * blkdev_nr_zones - Get number of zones
  83  * @bdev:       Target block device
  84  *
  85  * Description:
  86  *    Return the total number of zones of a zoned block device.
  87  *    For a regular block device, the number of zones is always 0.
  88  */
  89 unsigned int blkdev_nr_zones(struct block_device *bdev)
  90 {
  91         struct request_queue *q = bdev_get_queue(bdev);
  92
  93         if (!blk_queue_is_zoned(q))
  94                 return 0;
  95
  96         return __blkdev_nr_zones(q, bdev->bd_part->nr_sects);
  97 }
  98 EXPORT_SYMBOL_GPL(blkdev_nr_zones);
  99
 100 /*
 101  * Check that a zone report belongs to this partition, and if yes, fix its start
 102  * sector and write pointer and return true. Return false otherwise.
 103  */
 104 static bool blkdev_report_zone(struct block_device *bdev, struct blk_zone *rep)
 105 {
 106         sector_t offset = get_start_sect(bdev);
 107
 108         if (rep->start < offset)
 109                 return false;
 110
 111         rep->start -= offset;
 112         if (rep->start + rep->len > bdev->bd_part->nr_sects)
 113                 return false;
 114
 115         if (rep->type == BLK_ZONE_TYPE_CONVENTIONAL)
 116                 rep->wp = rep->start + rep->len;
 117         else
 118                 rep->wp -= offset;
 119         return true;
 120 }
 121
 122 static int blk_report_zones(struct gendisk *disk, sector_t sector,
 123                             struct blk_zone *zones, unsigned int *nr_zones)
 124 {
 125         struct request_queue *q = disk->queue;
 126         unsigned int z = 0, n, nrz = *nr_zones;
 127         sector_t capacity = get_capacity(disk);
 128         int ret;
 129
 130         while (z < nrz && sector < capacity) {
 131                 n = nrz - z;
 132                 ret = disk->fops->report_zones(disk, sector, &zones[z], &n);
 133                 if (ret)
 134                         return ret;
 135                 if (!n)
 136                         break;
 137                 sector += blk_queue_zone_sectors(q) * n;
 138                 z += n;
 139         }
 140
 141         WARN_ON(z > *nr_zones);
 142         *nr_zones = z;
 143
 144         return 0;
 145 }
 146
 147 /**
 148  * blkdev_report_zones - Get zones information
 149  * @bdev:       Target block device
 150  * @sector:     Sector from which to report zones
 151  * @zones:      Array of zone structures where to return the zones information
 152  * @nr_zones:   Number of zone structures in the zone array
 153  *
 154  * Description:
 155  *    Get zone information starting from the zone containing @sector.
 156  *    The number of zone information reported may be less than the number
 157  *    requested by @nr_zones. The number of zones actually reported is
 158  *    returned in @nr_zones.
 159  *    The caller must use memalloc_noXX_save/restore() calls to control
 160  *    memory allocations done within this function (zone array and command
 161  *    buffer allocation by the device driver).
 162  */
 163 int blkdev_report_zones(struct block_device *bdev, sector_t sector,
 164                         struct blk_zone *zones, unsigned int *nr_zones)
 165 {
 166         struct request_queue *q = bdev_get_queue(bdev);
 167         unsigned int i, nrz;
 168         int ret;
 169
 170         if (!blk_queue_is_zoned(q))
 171                 return -EOPNOTSUPP;
 172
 173         /*
 174          * A block device that advertized itself as zoned must have a
 175          * report_zones method. If it does not have one defined, the device
 176          * driver has a bug. So warn about that.
 177          */
 178         if (WARN_ON_ONCE(!bdev->bd_disk->fops->report_zones))
 179                 return -EOPNOTSUPP;
 180
 181         if (!*nr_zones || sector >= bdev->bd_part->nr_sects) {
 182                 *nr_zones = 0;
 183                 return 0;
 184         }
 185
 186         nrz = min(*nr_zones,
 187                   __blkdev_nr_zones(q, bdev->bd_part->nr_sects - sector));
 188         ret = blk_report_zones(bdev->bd_disk, get_start_sect(bdev) + sector,
 189                                zones, &nrz);
 190         if (ret)
 191                 return ret;
 192
 193         for (i = 0; i < nrz; i++) {
 194                 if (!blkdev_report_zone(bdev, zones))
 195                         break;
 196                 zones++;
 197         }
 198
 199         *nr_zones = i;
 200
 201         return 0;
 202 }
 203 EXPORT_SYMBOL_GPL(blkdev_report_zones);
 204
 205 /**
 206  * blkdev_reset_zones - Reset zones write pointer
 207  * @bdev:       Target block device
 208  * @sector:     Start sector of the first zone to reset
 209  * @nr_sectors: Number of sectors, at least the length of one zone
 210  * @gfp_mask:   Memory allocation flags (for bio_alloc)
 211  *
 212  * Description:
 213  *    Reset the write pointer of the zones contained in the range
 214  *    @sector..@sector+@nr_sectors. Specifying the entire disk sector range
 215  *    is valid, but the specified range should not contain conventional zones.
 216  */
 217 int blkdev_reset_zones(struct block_device *bdev,
 218                        sector_t sector, sector_t nr_sectors,
 219                        gfp_t gfp_mask)
 220 {
 221         struct request_queue *q = bdev_get_queue(bdev);
 222         sector_t zone_sectors;
 223         sector_t end_sector = sector + nr_sectors;
 224         struct bio *bio = NULL;
 225         struct blk_plug plug;
 226         int ret;
 227
 228         if (!blk_queue_is_zoned(q))
 229                 return -EOPNOTSUPP;
 230
 231         if (bdev_read_only(bdev))
 232                 return -EPERM;
 233
 234         if (!nr_sectors || end_sector > bdev->bd_part->nr_sects)
 235                 /* Out of range */
 236                 return -EINVAL;
 237
 238         /* Check alignment (handle eventual smaller last zone) */
 239         zone_sectors = blk_queue_zone_sectors(q);
 240         if (sector & (zone_sectors - 1))
 241                 return -EINVAL;
 242
 243         if ((nr_sectors & (zone_sectors - 1)) &&
 244             end_sector != bdev->bd_part->nr_sects)
 245                 return -EINVAL;
 246
 247         blk_start_plug(&plug);
 248         while (sector < end_sector) {
 249
 250                 bio = blk_next_bio(bio, 0, gfp_mask);
 251                 bio->bi_iter.bi_sector = sector;
 252                 bio_set_dev(bio, bdev);
 253                 bio_set_op_attrs(bio, REQ_OP_ZONE_RESET, 0);
 254
 255                 sector += zone_sectors;
 256
 257                 /* This may take a while, so be nice to others */
 258                 cond_resched();
 259
 260         }
 261
 262         ret = submit_bio_wait(bio);
 263         bio_put(bio);
 264
 265         blk_finish_plug(&plug);
 266
 267         return ret;
 268 }
 269 EXPORT_SYMBOL_GPL(blkdev_reset_zones);
 270
 271 /*
 272  * BLKREPORTZONE ioctl processing.
 273  * Called from blkdev_ioctl.
 274  */
 275 int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
 276                               unsigned int cmd, unsigned long arg)
 277 {
 278         void __user *argp = (void __user *)arg;
 279         struct request_queue *q;
 280         struct blk_zone_report rep;
 281         struct blk_zone *zones;
 282         int ret;
 283
 284         if (!argp)
 285                 return -EINVAL;
 286
 287         q = bdev_get_queue(bdev);
 288         if (!q)
 289                 return -ENXIO;
 290
 291         if (!blk_queue_is_zoned(q))
 292                 return -ENOTTY;
 293
 294         if (!capable(CAP_SYS_ADMIN))
 295                 return -EACCES;
 296
 297         if (copy_from_user(&rep, argp, sizeof(struct blk_zone_report)))
 298                 return -EFAULT;
 299
 300         if (!rep.nr_zones)
 301                 return -EINVAL;
 302
 303         rep.nr_zones = min(blkdev_nr_zones(bdev), rep.nr_zones);
 304
 305         zones = kvmalloc_array(rep.nr_zones, sizeof(struct blk_zone),
 306                                GFP_KERNEL | __GFP_ZERO);
 307         if (!zones)
 308                 return -ENOMEM;
 309
 310         ret = blkdev_report_zones(bdev, rep.sector, zones, &rep.nr_zones);
 311         if (ret)
 312                 goto out;
 313
 314         if (copy_to_user(argp, &rep, sizeof(struct blk_zone_report))) {
 315                 ret = -EFAULT;
 316                 goto out;
 317         }
 318
 319         if (rep.nr_zones) {
 320                 if (copy_to_user(argp + sizeof(struct blk_zone_report), zones,
 321                                  sizeof(struct blk_zone) * rep.nr_zones))
 322                         ret = -EFAULT;
 323         }
 324
 325  out:
 326         kvfree(zones);
 327
 328         return ret;
 329 }
 330
 331 /*
 332  * BLKRESETZONE ioctl processing.
 333  * Called from blkdev_ioctl.
 334  */
 335 int blkdev_reset_zones_ioctl(struct block_device *bdev, fmode_t mode,
 336                              unsigned int cmd, unsigned long arg)
 337 {
 338         void __user *argp = (void __user *)arg;
 339         struct request_queue *q;
 340         struct blk_zone_range zrange;
 341
 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
 361         return blkdev_reset_zones(bdev, zrange.sector, zrange.nr_sectors,
 362                                   GFP_KERNEL);
 363 }
 364
 365 static inline unsigned long *blk_alloc_zone_bitmap(int node,
 366                                                    unsigned int nr_zones)
 367 {
 368         return kcalloc_node(BITS_TO_LONGS(nr_zones), sizeof(unsigned long),
 369                             GFP_NOIO, node);
 370 }
 371
 372 /*
 373  * Allocate an array of struct blk_zone to get nr_zones zone information.
 374  * The allocated array may be smaller than nr_zones.
 375  */
 376 static struct blk_zone *blk_alloc_zones(unsigned int *nr_zones)
 377 {
 378         struct blk_zone *zones;
 379         size_t nrz = min(*nr_zones, BLK_ZONED_REPORT_MAX_ZONES);
 380
 381         /*
 382          * GFP_KERNEL here is meaningless as the caller task context has
 383          * the PF_MEMALLOC_NOIO flag set in blk_revalidate_disk_zones()
 384          * with memalloc_noio_save().
 385          */
 386         zones = kvcalloc(nrz, sizeof(struct blk_zone), GFP_KERNEL);
 387         if (!zones) {
 388                 *nr_zones = 0;
 389                 return NULL;
 390         }
 391
 392         *nr_zones = nrz;
 393
 394         return zones;
 395 }
 396
 397 void blk_queue_free_zone_bitmaps(struct request_queue *q)
 398 {
 399         kfree(q->seq_zones_bitmap);
 400         q->seq_zones_bitmap = NULL;
 401         kfree(q->seq_zones_wlock);
 402         q->seq_zones_wlock = NULL;
 403 }
 404
 405 /**
 406  * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
 407  * @disk:       Target disk
 408  *
 409  * Helper function for low-level device drivers to (re) allocate and initialize
 410  * a disk request queue zone bitmaps. This functions should normally be called
 411  * within the disk ->revalidate method. For BIO based queues, no zone bitmap
 412  * is allocated.
 413  */
 414 int blk_revalidate_disk_zones(struct gendisk *disk)
 415 {
 416         struct request_queue *q = disk->queue;
 417         unsigned int nr_zones = __blkdev_nr_zones(q, get_capacity(disk));
 418         unsigned long *seq_zones_wlock = NULL, *seq_zones_bitmap = NULL;
 419         unsigned int i, rep_nr_zones = 0, z = 0, nrz;
 420         struct blk_zone *zones = NULL;
 421         unsigned int noio_flag;
 422         sector_t sector = 0;
 423         int ret = 0;
 424
 425         /*
 426          * BIO based queues do not use a scheduler so only q->nr_zones
 427          * needs to be updated so that the sysfs exposed value is correct.
 428          */
 429         if (!queue_is_mq(q)) {
 430                 q->nr_zones = nr_zones;
 431                 return 0;
 432         }
 433
 434         /*
 435          * Ensure that all memory allocations in this context are done as
 436          * if GFP_NOIO was specified.
 437          */
 438         noio_flag = memalloc_noio_save();
 439
 440         if (!blk_queue_is_zoned(q) || !nr_zones) {
 441                 nr_zones = 0;
 442                 goto update;
 443         }
 444
 445         /* Allocate bitmaps */
 446         ret = -ENOMEM;
 447         seq_zones_wlock = blk_alloc_zone_bitmap(q->node, nr_zones);
 448         if (!seq_zones_wlock)
 449                 goto out;
 450         seq_zones_bitmap = blk_alloc_zone_bitmap(q->node, nr_zones);
 451         if (!seq_zones_bitmap)
 452                 goto out;
 453
 454         /* Get zone information and initialize seq_zones_bitmap */
 455         rep_nr_zones = nr_zones;
 456         zones = blk_alloc_zones(&rep_nr_zones);
 457         if (!zones)
 458                 goto out;
 459
 460         while (z < nr_zones) {
 461                 nrz = min(nr_zones - z, rep_nr_zones);
 462                 ret = blk_report_zones(disk, sector, zones, &nrz);
 463                 if (ret)
 464                         goto out;
 465                 if (!nrz)
 466                         break;
 467                 for (i = 0; i < nrz; i++) {
 468                         if (zones[i].type != BLK_ZONE_TYPE_CONVENTIONAL)
 469                                 set_bit(z, seq_zones_bitmap);
 470                         z++;
 471                 }
 472                 sector += nrz * blk_queue_zone_sectors(q);
 473         }
 474
 475         if (WARN_ON(z != nr_zones)) {
 476                 ret = -EIO;
 477                 goto out;
 478         }
 479
 480 update:
 481         /*
 482          * Install the new bitmaps, making sure the queue is stopped and
 483          * all I/Os are completed (i.e. a scheduler is not referencing the
 484          * bitmaps).
 485          */
 486         blk_mq_freeze_queue(q);
 487         q->nr_zones = nr_zones;
 488         swap(q->seq_zones_wlock, seq_zones_wlock);
 489         swap(q->seq_zones_bitmap, seq_zones_bitmap);
 490         blk_mq_unfreeze_queue(q);
 491
 492 out:
 493         memalloc_noio_restore(noio_flag);
 494
 495         kvfree(zones);
 496         kfree(seq_zones_wlock);
 497         kfree(seq_zones_bitmap);
 498
 499         if (ret) {
 500                 pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
 501                 blk_mq_freeze_queue(q);
 502                 blk_queue_free_zone_bitmaps(q);
 503                 blk_mq_unfreeze_queue(q);
 504         }
 505
 506         return ret;
 507 }
 508 EXPORT_SYMBOL_GPL(blk_revalidate_disk_zones);
 509