block/partition-generic.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  Code extracted from drivers/block/genhd.c
   4  *  Copyright (C) 1991-1998  Linus Torvalds
   5  *  Re-organised Feb 1998 Russell King
   6  *
   7  *  We now have independent partition support from the
   8  *  block drivers, which allows all the partition code to
   9  *  be grouped in one location, and it to be mostly self
  10  *  contained.
  11  */
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/fs.h>
  16 #include <linux/slab.h>
  17 #include <linux/kmod.h>
  18 #include <linux/ctype.h>
  19 #include <linux/genhd.h>
  20 #include <linux/blktrace_api.h>
  21
  22 #include "partitions/check.h"
  23
  24 #ifdef CONFIG_BLK_DEV_MD
  25 extern void md_autodetect_dev(dev_t dev);
  26 #endif
  27
  28 /*
  29  * disk_name() is used by partition check code and the genhd driver.
  30  * It formats the devicename of the indicated disk into
  31  * the supplied buffer (of size at least 32), and returns
  32  * a pointer to that same buffer (for convenience).
  33  */
  34
  35 char *disk_name(struct gendisk *hd, int partno, char *buf)
  36 {
  37         if (!partno)
  38                 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
  39         else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
  40                 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
  41         else
  42                 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
  43
  44         return buf;
  45 }
  46
  47 const char *bdevname(struct block_device *bdev, char *buf)
  48 {
  49         return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
  50 }
  51
  52 EXPORT_SYMBOL(bdevname);
  53
  54 const char *bio_devname(struct bio *bio, char *buf)
  55 {
  56         return disk_name(bio->bi_disk, bio->bi_partno, buf);
  57 }
  58 EXPORT_SYMBOL(bio_devname);
  59
  60 /*
  61  * There's very little reason to use this, you should really
  62  * have a struct block_device just about everywhere and use
  63  * bdevname() instead.
  64  */
  65 const char *__bdevname(dev_t dev, char *buffer)
  66 {
  67         scnprintf(buffer, BDEVNAME_SIZE, "unknown-block(%u,%u)",
  68                                 MAJOR(dev), MINOR(dev));
  69         return buffer;
  70 }
  71
  72 EXPORT_SYMBOL(__bdevname);
  73
  74 static ssize_t part_partition_show(struct device *dev,
  75                                    struct device_attribute *attr, char *buf)
  76 {
  77         struct hd_struct *p = dev_to_part(dev);
  78
  79         return sprintf(buf, "%d\n", p->partno);
  80 }
  81
  82 static ssize_t part_start_show(struct device *dev,
  83                                struct device_attribute *attr, char *buf)
  84 {
  85         struct hd_struct *p = dev_to_part(dev);
  86
  87         return sprintf(buf, "%llu\n",(unsigned long long)p->start_sect);
  88 }
  89
  90 ssize_t part_size_show(struct device *dev,
  91                        struct device_attribute *attr, char *buf)
  92 {
  93         struct hd_struct *p = dev_to_part(dev);
  94         return sprintf(buf, "%llu\n",(unsigned long long)part_nr_sects_read(p));
  95 }
  96
  97 static ssize_t part_ro_show(struct device *dev,
  98                             struct device_attribute *attr, char *buf)
  99 {
 100         struct hd_struct *p = dev_to_part(dev);
 101         return sprintf(buf, "%d\n", p->policy ? 1 : 0);
 102 }
 103
 104 static ssize_t part_alignment_offset_show(struct device *dev,
 105                                           struct device_attribute *attr, char *buf)
 106 {
 107         struct hd_struct *p = dev_to_part(dev);
 108         return sprintf(buf, "%llu\n", (unsigned long long)p->alignment_offset);
 109 }
 110
 111 static ssize_t part_discard_alignment_show(struct device *dev,
 112                                            struct device_attribute *attr, char *buf)
 113 {
 114         struct hd_struct *p = dev_to_part(dev);
 115         return sprintf(buf, "%u\n", p->discard_alignment);
 116 }
 117
 118 ssize_t part_stat_show(struct device *dev,
 119                        struct device_attribute *attr, char *buf)
 120 {
 121         struct hd_struct *p = dev_to_part(dev);
 122         struct request_queue *q = part_to_disk(p)->queue;
 123         unsigned int inflight;
 124
 125         inflight = part_in_flight(q, p);
 126         return sprintf(buf,
 127                 "%8lu %8lu %8llu %8u "
 128                 "%8lu %8lu %8llu %8u "
 129                 "%8u %8u %8u "
 130                 "%8lu %8lu %8llu %8u"
 131                 "\n",
 132                 part_stat_read(p, ios[STAT_READ]),
 133                 part_stat_read(p, merges[STAT_READ]),
 134                 (unsigned long long)part_stat_read(p, sectors[STAT_READ]),
 135                 (unsigned int)part_stat_read_msecs(p, STAT_READ),
 136                 part_stat_read(p, ios[STAT_WRITE]),
 137                 part_stat_read(p, merges[STAT_WRITE]),
 138                 (unsigned long long)part_stat_read(p, sectors[STAT_WRITE]),
 139                 (unsigned int)part_stat_read_msecs(p, STAT_WRITE),
 140                 inflight,
 141                 jiffies_to_msecs(part_stat_read(p, io_ticks)),
 142                 jiffies_to_msecs(part_stat_read(p, time_in_queue)),
 143                 part_stat_read(p, ios[STAT_DISCARD]),
 144                 part_stat_read(p, merges[STAT_DISCARD]),
 145                 (unsigned long long)part_stat_read(p, sectors[STAT_DISCARD]),
 146                 (unsigned int)part_stat_read_msecs(p, STAT_DISCARD));
 147 }
 148
 149 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
 150                            char *buf)
 151 {
 152         struct hd_struct *p = dev_to_part(dev);
 153         struct request_queue *q = part_to_disk(p)->queue;
 154         unsigned int inflight[2];
 155
 156         part_in_flight_rw(q, p, inflight);
 157         return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
 158 }
 159
 160 #ifdef CONFIG_FAIL_MAKE_REQUEST
 161 ssize_t part_fail_show(struct device *dev,
 162                        struct device_attribute *attr, char *buf)
 163 {
 164         struct hd_struct *p = dev_to_part(dev);
 165
 166         return sprintf(buf, "%d\n", p->make_it_fail);
 167 }
 168
 169 ssize_t part_fail_store(struct device *dev,
 170                         struct device_attribute *attr,
 171                         const char *buf, size_t count)
 172 {
 173         struct hd_struct *p = dev_to_part(dev);
 174         int i;
 175
 176         if (count > 0 && sscanf(buf, "%d", &i) > 0)
 177                 p->make_it_fail = (i == 0) ? 0 : 1;
 178
 179         return count;
 180 }
 181 #endif
 182
 183 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
 184 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
 185 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
 186 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
 187 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
 188 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
 189 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
 190 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
 191 #ifdef CONFIG_FAIL_MAKE_REQUEST
 192 static struct device_attribute dev_attr_fail =
 193         __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
 194 #endif
 195
 196 static struct attribute *part_attrs[] = {
 197         &dev_attr_partition.attr,
 198         &dev_attr_start.attr,
 199         &dev_attr_size.attr,
 200         &dev_attr_ro.attr,
 201         &dev_attr_alignment_offset.attr,
 202         &dev_attr_discard_alignment.attr,
 203         &dev_attr_stat.attr,
 204         &dev_attr_inflight.attr,
 205 #ifdef CONFIG_FAIL_MAKE_REQUEST
 206         &dev_attr_fail.attr,
 207 #endif
 208         NULL
 209 };
 210
 211 static struct attribute_group part_attr_group = {
 212         .attrs = part_attrs,
 213 };
 214
 215 static const struct attribute_group *part_attr_groups[] = {
 216         &part_attr_group,
 217 #ifdef CONFIG_BLK_DEV_IO_TRACE
 218         &blk_trace_attr_group,
 219 #endif
 220         NULL
 221 };
 222
 223 static void part_release(struct device *dev)
 224 {
 225         struct hd_struct *p = dev_to_part(dev);
 226         blk_free_devt(dev->devt);
 227         hd_free_part(p);
 228         kfree(p);
 229 }
 230
 231 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
 232 {
 233         struct hd_struct *part = dev_to_part(dev);
 234
 235         add_uevent_var(env, "PARTN=%u", part->partno);
 236         if (part->info && part->info->volname[0])
 237                 add_uevent_var(env, "PARTNAME=%s", part->info->volname);
 238         return 0;
 239 }
 240
 241 struct device_type part_type = {
 242         .name           = "partition",
 243         .groups         = part_attr_groups,
 244         .release        = part_release,
 245         .uevent         = part_uevent,
 246 };
 247
 248 static void delete_partition_work_fn(struct work_struct *work)
 249 {
 250         struct hd_struct *part = container_of(to_rcu_work(work), struct hd_struct,
 251                                         rcu_work);
 252
 253         part->start_sect = 0;
 254         part->nr_sects = 0;
 255         part_stat_set_all(part, 0);
 256         put_device(part_to_dev(part));
 257 }
 258
 259 void __delete_partition(struct percpu_ref *ref)
 260 {
 261         struct hd_struct *part = container_of(ref, struct hd_struct, ref);
 262         INIT_RCU_WORK(&part->rcu_work, delete_partition_work_fn);
 263         queue_rcu_work(system_wq, &part->rcu_work);
 264 }
 265
 266 /*
 267  * Must be called either with bd_mutex held, before a disk can be opened or
 268  * after all disk users are gone.
 269  */
 270 void delete_partition(struct gendisk *disk, int partno)
 271 {
 272         struct disk_part_tbl *ptbl =
 273                 rcu_dereference_protected(disk->part_tbl, 1);
 274         struct hd_struct *part;
 275
 276         if (partno >= ptbl->len)
 277                 return;
 278
 279         part = rcu_dereference_protected(ptbl->part[partno], 1);
 280         if (!part)
 281                 return;
 282
 283         rcu_assign_pointer(ptbl->part[partno], NULL);
 284         rcu_assign_pointer(ptbl->last_lookup, NULL);
 285         kobject_put(part->holder_dir);
 286         device_del(part_to_dev(part));
 287
 288         /*
 289          * Remove gendisk pointer from idr so that it cannot be looked up
 290          * while RCU period before freeing gendisk is running to prevent
 291          * use-after-free issues. Note that the device number stays
 292          * "in-use" until we really free the gendisk.
 293          */
 294         blk_invalidate_devt(part_devt(part));
 295         hd_struct_kill(part);
 296 }
 297
 298 static ssize_t whole_disk_show(struct device *dev,
 299                                struct device_attribute *attr, char *buf)
 300 {
 301         return 0;
 302 }
 303 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
 304
 305 /*
 306  * Must be called either with bd_mutex held, before a disk can be opened or
 307  * after all disk users are gone.
 308  */
 309 struct hd_struct *add_partition(struct gendisk *disk, int partno,
 310                                 sector_t start, sector_t len, int flags,
 311                                 struct partition_meta_info *info)
 312 {
 313         struct hd_struct *p;
 314         dev_t devt = MKDEV(0, 0);
 315         struct device *ddev = disk_to_dev(disk);
 316         struct device *pdev;
 317         struct disk_part_tbl *ptbl;
 318         const char *dname;
 319         int err;
 320
 321         err = disk_expand_part_tbl(disk, partno);
 322         if (err)
 323                 return ERR_PTR(err);
 324         ptbl = rcu_dereference_protected(disk->part_tbl, 1);
 325
 326         if (ptbl->part[partno])
 327                 return ERR_PTR(-EBUSY);
 328
 329         p = kzalloc(sizeof(*p), GFP_KERNEL);
 330         if (!p)
 331                 return ERR_PTR(-EBUSY);
 332
 333         if (!init_part_stats(p)) {
 334                 err = -ENOMEM;
 335                 goto out_free;
 336         }
 337
 338         seqcount_init(&p->nr_sects_seq);
 339         pdev = part_to_dev(p);
 340
 341         p->start_sect = start;
 342         p->alignment_offset =
 343                 queue_limit_alignment_offset(&disk->queue->limits, start);
 344         p->discard_alignment =
 345                 queue_limit_discard_alignment(&disk->queue->limits, start);
 346         p->nr_sects = len;
 347         p->partno = partno;
 348         p->policy = get_disk_ro(disk);
 349
 350         if (info) {
 351                 struct partition_meta_info *pinfo = alloc_part_info(disk);
 352                 if (!pinfo) {
 353                         err = -ENOMEM;
 354                         goto out_free_stats;
 355                 }
 356                 memcpy(pinfo, info, sizeof(*info));
 357                 p->info = pinfo;
 358         }
 359
 360         dname = dev_name(ddev);
 361         if (isdigit(dname[strlen(dname) - 1]))
 362                 dev_set_name(pdev, "%sp%d", dname, partno);
 363         else
 364                 dev_set_name(pdev, "%s%d", dname, partno);
 365
 366         device_initialize(pdev);
 367         pdev->class = &block_class;
 368         pdev->type = &part_type;
 369         pdev->parent = ddev;
 370
 371         err = blk_alloc_devt(p, &devt);
 372         if (err)
 373                 goto out_free_info;
 374         pdev->devt = devt;
 375
 376         /* delay uevent until 'holders' subdir is created */
 377         dev_set_uevent_suppress(pdev, 1);
 378         err = device_add(pdev);
 379         if (err)
 380                 goto out_put;
 381
 382         err = -ENOMEM;
 383         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
 384         if (!p->holder_dir)
 385                 goto out_del;
 386
 387         dev_set_uevent_suppress(pdev, 0);
 388         if (flags & ADDPART_FLAG_WHOLEDISK) {
 389                 err = device_create_file(pdev, &dev_attr_whole_disk);
 390                 if (err)
 391                         goto out_del;
 392         }
 393
 394         err = hd_ref_init(p);
 395         if (err) {
 396                 if (flags & ADDPART_FLAG_WHOLEDISK)
 397                         goto out_remove_file;
 398                 goto out_del;
 399         }
 400
 401         /* everything is up and running, commence */
 402         rcu_assign_pointer(ptbl->part[partno], p);
 403
 404         /* suppress uevent if the disk suppresses it */
 405         if (!dev_get_uevent_suppress(ddev))
 406                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
 407         return p;
 408
 409 out_free_info:
 410         free_part_info(p);
 411 out_free_stats:
 412         free_part_stats(p);
 413 out_free:
 414         kfree(p);
 415         return ERR_PTR(err);
 416 out_remove_file:
 417         device_remove_file(pdev, &dev_attr_whole_disk);
 418 out_del:
 419         kobject_put(p->holder_dir);
 420         device_del(pdev);
 421 out_put:
 422         put_device(pdev);
 423         return ERR_PTR(err);
 424 }
 425
 426 static bool disk_unlock_native_capacity(struct gendisk *disk)
 427 {
 428         const struct block_device_operations *bdops = disk->fops;
 429
 430         if (bdops->unlock_native_capacity &&
 431             !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
 432                 printk(KERN_CONT "enabling native capacity\n");
 433                 bdops->unlock_native_capacity(disk);
 434                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
 435                 return true;
 436         } else {
 437                 printk(KERN_CONT "truncated\n");
 438                 return false;
 439         }
 440 }
 441
 442 static int drop_partitions(struct gendisk *disk, struct block_device *bdev)
 443 {
 444         struct disk_part_iter piter;
 445         struct hd_struct *part;
 446         int res;
 447
 448         if (bdev->bd_part_count || bdev->bd_super)
 449                 return -EBUSY;
 450         res = invalidate_partition(disk, 0);
 451         if (res)
 452                 return res;
 453
 454         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 455         while ((part = disk_part_iter_next(&piter)))
 456                 delete_partition(disk, part->partno);
 457         disk_part_iter_exit(&piter);
 458
 459         return 0;
 460 }
 461
 462 static bool blk_add_partition(struct gendisk *disk, struct block_device *bdev,
 463                 struct parsed_partitions *state, int p)
 464 {
 465         sector_t size = state->parts[p].size;
 466         sector_t from = state->parts[p].from;
 467         struct hd_struct *part;
 468
 469         if (!size)
 470                 return true;
 471
 472         if (from >= get_capacity(disk)) {
 473                 printk(KERN_WARNING
 474                        "%s: p%d start %llu is beyond EOD, ",
 475                        disk->disk_name, p, (unsigned long long) from);
 476                 if (disk_unlock_native_capacity(disk))
 477                         return false;
 478                 return true;
 479         }
 480
 481         if (from + size > get_capacity(disk)) {
 482                 printk(KERN_WARNING
 483                        "%s: p%d size %llu extends beyond EOD, ",
 484                        disk->disk_name, p, (unsigned long long) size);
 485
 486                 if (disk_unlock_native_capacity(disk))
 487                         return false;
 488
 489                 /*
 490                  * We can not ignore partitions of broken tables created by for
 491                  * example camera firmware, but we limit them to the end of the
 492                  * disk to avoid creating invalid block devices.
 493                  */
 494                 size = get_capacity(disk) - from;
 495         }
 496
 497         part = add_partition(disk, p, from, size, state->parts[p].flags,
 498                              &state->parts[p].info);
 499         if (IS_ERR(part)) {
 500                 printk(KERN_ERR " %s: p%d could not be added: %ld\n",
 501                        disk->disk_name, p, -PTR_ERR(part));
 502                 return true;
 503         }
 504
 505 #ifdef CONFIG_BLK_DEV_MD
 506         if (state->parts[p].flags & ADDPART_FLAG_RAID)
 507                 md_autodetect_dev(part_to_dev(part)->devt);
 508 #endif
 509         return true;
 510 }
 511
 512 static int blk_add_partitions(struct gendisk *disk, struct block_device *bdev)
 513 {
 514         struct parsed_partitions *state;
 515         int ret = -EAGAIN, p, highest;
 516
 517         state = check_partition(disk, bdev);
 518         if (!state)
 519                 return 0;
 520         if (IS_ERR(state)) {
 521                 /*
 522                  * I/O error reading the partition table.  If we tried to read
 523                  * beyond EOD, retry after unlocking the native capacity.
 524                  */
 525                 if (PTR_ERR(state) == -ENOSPC) {
 526                         printk(KERN_WARNING "%s: partition table beyond EOD, ",
 527                                disk->disk_name);
 528                         if (disk_unlock_native_capacity(disk))
 529                                 return -EAGAIN;
 530                 }
 531                 return -EIO;
 532         }
 533
 534         /*
 535          * Partitions are not supported on zoned block devices.
 536          */
 537         if (bdev_is_zoned(bdev)) {
 538                 pr_warn("%s: ignoring partition table on zoned block device\n",
 539                         disk->disk_name);
 540                 ret = 0;
 541                 goto out_free_state;
 542         }
 543
 544         /*
 545          * If we read beyond EOD, try unlocking native capacity even if the
 546          * partition table was successfully read as we could be missing some
 547          * partitions.
 548          */
 549         if (state->access_beyond_eod) {
 550                 printk(KERN_WARNING
 551                        "%s: partition table partially beyond EOD, ",
 552                        disk->disk_name);
 553                 if (disk_unlock_native_capacity(disk))
 554                         goto out_free_state;
 555         }
 556
 557         /* tell userspace that the media / partition table may have changed */
 558         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 559
 560         /*
 561          * Detect the highest partition number and preallocate disk->part_tbl.
 562          * This is an optimization and not strictly necessary.
 563          */
 564         for (p = 1, highest = 0; p < state->limit; p++)
 565                 if (state->parts[p].size)
 566                         highest = p;
 567         disk_expand_part_tbl(disk, highest);
 568
 569         for (p = 1; p < state->limit; p++)
 570                 if (!blk_add_partition(disk, bdev, state, p))
 571                         goto out_free_state;
 572
 573         ret = 0;
 574 out_free_state:
 575         free_partitions(state);
 576         return ret;
 577 }
 578
 579 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
 580 {
 581         int ret;
 582
 583 rescan:
 584         ret = drop_partitions(disk, bdev);
 585         if (ret)
 586                 return ret;
 587
 588         if (disk->fops->revalidate_disk)
 589                 disk->fops->revalidate_disk(disk);
 590         check_disk_size_change(disk, bdev, true);
 591         bdev->bd_invalidated = 0;
 592
 593         if (!get_capacity(disk))
 594                 return 0;
 595
 596         ret = blk_add_partitions(disk, bdev);
 597         if (ret == -EAGAIN)
 598                 goto rescan;
 599         return ret;
 600 }
 601
 602 int invalidate_partitions(struct gendisk *disk, struct block_device *bdev)
 603 {
 604         int res;
 605
 606         if (!bdev->bd_invalidated)
 607                 return 0;
 608
 609         res = drop_partitions(disk, bdev);
 610         if (res)
 611                 return res;
 612
 613         set_capacity(disk, 0);
 614         check_disk_size_change(disk, bdev, false);
 615         bdev->bd_invalidated = 0;
 616         /* tell userspace that the media / partition table may have changed */
 617         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 618
 619         return 0;
 620 }
 621
 622 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
 623 {
 624         struct address_space *mapping = bdev->bd_inode->i_mapping;
 625         struct page *page;
 626
 627         page = read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_SHIFT-9)), NULL);
 628         if (!IS_ERR(page)) {
 629                 if (PageError(page))
 630                         goto fail;
 631                 p->v = page;
 632                 return (unsigned char *)page_address(page) +  ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << 9);
 633 fail:
 634                 put_page(page);
 635         }
 636         p->v = NULL;
 637         return NULL;
 638 }
 639
 640 EXPORT_SYMBOL(read_dev_sector);