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                 "%8lu %8u"
 132                 "\n",
 133                 part_stat_read(p, ios[STAT_READ]),
 134                 part_stat_read(p, merges[STAT_READ]),
 135                 (unsigned long long)part_stat_read(p, sectors[STAT_READ]),
 136                 (unsigned int)part_stat_read_msecs(p, STAT_READ),
 137                 part_stat_read(p, ios[STAT_WRITE]),
 138                 part_stat_read(p, merges[STAT_WRITE]),
 139                 (unsigned long long)part_stat_read(p, sectors[STAT_WRITE]),
 140                 (unsigned int)part_stat_read_msecs(p, STAT_WRITE),
 141                 inflight,
 142                 jiffies_to_msecs(part_stat_read(p, io_ticks)),
 143                 jiffies_to_msecs(part_stat_read(p, time_in_queue)),
 144                 part_stat_read(p, ios[STAT_DISCARD]),
 145                 part_stat_read(p, merges[STAT_DISCARD]),
 146                 (unsigned long long)part_stat_read(p, sectors[STAT_DISCARD]),
 147                 (unsigned int)part_stat_read_msecs(p, STAT_DISCARD),
 148                 part_stat_read(p, ios[STAT_FLUSH]),
 149                 (unsigned int)part_stat_read_msecs(p, STAT_FLUSH));
 150 }
 151
 152 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
 153                            char *buf)
 154 {
 155         struct hd_struct *p = dev_to_part(dev);
 156         struct request_queue *q = part_to_disk(p)->queue;
 157         unsigned int inflight[2];
 158
 159         part_in_flight_rw(q, p, inflight);
 160         return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
 161 }
 162
 163 #ifdef CONFIG_FAIL_MAKE_REQUEST
 164 ssize_t part_fail_show(struct device *dev,
 165                        struct device_attribute *attr, char *buf)
 166 {
 167         struct hd_struct *p = dev_to_part(dev);
 168
 169         return sprintf(buf, "%d\n", p->make_it_fail);
 170 }
 171
 172 ssize_t part_fail_store(struct device *dev,
 173                         struct device_attribute *attr,
 174                         const char *buf, size_t count)
 175 {
 176         struct hd_struct *p = dev_to_part(dev);
 177         int i;
 178
 179         if (count > 0 && sscanf(buf, "%d", &i) > 0)
 180                 p->make_it_fail = (i == 0) ? 0 : 1;
 181
 182         return count;
 183 }
 184 #endif
 185
 186 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
 187 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
 188 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
 189 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
 190 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
 191 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
 192 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
 193 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
 194 #ifdef CONFIG_FAIL_MAKE_REQUEST
 195 static struct device_attribute dev_attr_fail =
 196         __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
 197 #endif
 198
 199 static struct attribute *part_attrs[] = {
 200         &dev_attr_partition.attr,
 201         &dev_attr_start.attr,
 202         &dev_attr_size.attr,
 203         &dev_attr_ro.attr,
 204         &dev_attr_alignment_offset.attr,
 205         &dev_attr_discard_alignment.attr,
 206         &dev_attr_stat.attr,
 207         &dev_attr_inflight.attr,
 208 #ifdef CONFIG_FAIL_MAKE_REQUEST
 209         &dev_attr_fail.attr,
 210 #endif
 211         NULL
 212 };
 213
 214 static struct attribute_group part_attr_group = {
 215         .attrs = part_attrs,
 216 };
 217
 218 static const struct attribute_group *part_attr_groups[] = {
 219         &part_attr_group,
 220 #ifdef CONFIG_BLK_DEV_IO_TRACE
 221         &blk_trace_attr_group,
 222 #endif
 223         NULL
 224 };
 225
 226 static void part_release(struct device *dev)
 227 {
 228         struct hd_struct *p = dev_to_part(dev);
 229         blk_free_devt(dev->devt);
 230         hd_free_part(p);
 231         kfree(p);
 232 }
 233
 234 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
 235 {
 236         struct hd_struct *part = dev_to_part(dev);
 237
 238         add_uevent_var(env, "PARTN=%u", part->partno);
 239         if (part->info && part->info->volname[0])
 240                 add_uevent_var(env, "PARTNAME=%s", part->info->volname);
 241         return 0;
 242 }
 243
 244 struct device_type part_type = {
 245         .name           = "partition",
 246         .groups         = part_attr_groups,
 247         .release        = part_release,
 248         .uevent         = part_uevent,
 249 };
 250
 251 static void delete_partition_work_fn(struct work_struct *work)
 252 {
 253         struct hd_struct *part = container_of(to_rcu_work(work), struct hd_struct,
 254                                         rcu_work);
 255
 256         part->start_sect = 0;
 257         part->nr_sects = 0;
 258         part_stat_set_all(part, 0);
 259         put_device(part_to_dev(part));
 260 }
 261
 262 void __delete_partition(struct percpu_ref *ref)
 263 {
 264         struct hd_struct *part = container_of(ref, struct hd_struct, ref);
 265         INIT_RCU_WORK(&part->rcu_work, delete_partition_work_fn);
 266         queue_rcu_work(system_wq, &part->rcu_work);
 267 }
 268
 269 /*
 270  * Must be called either with bd_mutex held, before a disk can be opened or
 271  * after all disk users are gone.
 272  */
 273 void delete_partition(struct gendisk *disk, int partno)
 274 {
 275         struct disk_part_tbl *ptbl =
 276                 rcu_dereference_protected(disk->part_tbl, 1);
 277         struct hd_struct *part;
 278
 279         if (partno >= ptbl->len)
 280                 return;
 281
 282         part = rcu_dereference_protected(ptbl->part[partno], 1);
 283         if (!part)
 284                 return;
 285
 286         rcu_assign_pointer(ptbl->part[partno], NULL);
 287         rcu_assign_pointer(ptbl->last_lookup, NULL);
 288         kobject_put(part->holder_dir);
 289         device_del(part_to_dev(part));
 290
 291         /*
 292          * Remove gendisk pointer from idr so that it cannot be looked up
 293          * while RCU period before freeing gendisk is running to prevent
 294          * use-after-free issues. Note that the device number stays
 295          * "in-use" until we really free the gendisk.
 296          */
 297         blk_invalidate_devt(part_devt(part));
 298         hd_struct_kill(part);
 299 }
 300
 301 static ssize_t whole_disk_show(struct device *dev,
 302                                struct device_attribute *attr, char *buf)
 303 {
 304         return 0;
 305 }
 306 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
 307
 308 /*
 309  * Must be called either with bd_mutex held, before a disk can be opened or
 310  * after all disk users are gone.
 311  */
 312 struct hd_struct *add_partition(struct gendisk *disk, int partno,
 313                                 sector_t start, sector_t len, int flags,
 314                                 struct partition_meta_info *info)
 315 {
 316         struct hd_struct *p;
 317         dev_t devt = MKDEV(0, 0);
 318         struct device *ddev = disk_to_dev(disk);
 319         struct device *pdev;
 320         struct disk_part_tbl *ptbl;
 321         const char *dname;
 322         int err;
 323
 324         /*
 325          * Partitions are not supported on zoned block devices that are used as
 326          * such.
 327          */
 328         switch (disk->queue->limits.zoned) {
 329         case BLK_ZONED_HM:
 330                 pr_warn("%s: partitions not supported on host managed zoned block device\n",
 331                         disk->disk_name);
 332                 return ERR_PTR(-ENXIO);
 333         case BLK_ZONED_HA:
 334                 pr_info("%s: disabling host aware zoned block device support due to partitions\n",
 335                         disk->disk_name);
 336                 disk->queue->limits.zoned = BLK_ZONED_NONE;
 337                 break;
 338         case BLK_ZONED_NONE:
 339                 break;
 340         }
 341
 342         err = disk_expand_part_tbl(disk, partno);
 343         if (err)
 344                 return ERR_PTR(err);
 345         ptbl = rcu_dereference_protected(disk->part_tbl, 1);
 346
 347         if (ptbl->part[partno])
 348                 return ERR_PTR(-EBUSY);
 349
 350         p = kzalloc(sizeof(*p), GFP_KERNEL);
 351         if (!p)
 352                 return ERR_PTR(-EBUSY);
 353
 354         if (!init_part_stats(p)) {
 355                 err = -ENOMEM;
 356                 goto out_free;
 357         }
 358
 359         seqcount_init(&p->nr_sects_seq);
 360         pdev = part_to_dev(p);
 361
 362         p->start_sect = start;
 363         p->alignment_offset =
 364                 queue_limit_alignment_offset(&disk->queue->limits, start);
 365         p->discard_alignment =
 366                 queue_limit_discard_alignment(&disk->queue->limits, start);
 367         p->nr_sects = len;
 368         p->partno = partno;
 369         p->policy = get_disk_ro(disk);
 370
 371         if (info) {
 372                 struct partition_meta_info *pinfo = alloc_part_info(disk);
 373                 if (!pinfo) {
 374                         err = -ENOMEM;
 375                         goto out_free_stats;
 376                 }
 377                 memcpy(pinfo, info, sizeof(*info));
 378                 p->info = pinfo;
 379         }
 380
 381         dname = dev_name(ddev);
 382         if (isdigit(dname[strlen(dname) - 1]))
 383                 dev_set_name(pdev, "%sp%d", dname, partno);
 384         else
 385                 dev_set_name(pdev, "%s%d", dname, partno);
 386
 387         device_initialize(pdev);
 388         pdev->class = &block_class;
 389         pdev->type = &part_type;
 390         pdev->parent = ddev;
 391
 392         err = blk_alloc_devt(p, &devt);
 393         if (err)
 394                 goto out_free_info;
 395         pdev->devt = devt;
 396
 397         /* delay uevent until 'holders' subdir is created */
 398         dev_set_uevent_suppress(pdev, 1);
 399         err = device_add(pdev);
 400         if (err)
 401                 goto out_put;
 402
 403         err = -ENOMEM;
 404         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
 405         if (!p->holder_dir)
 406                 goto out_del;
 407
 408         dev_set_uevent_suppress(pdev, 0);
 409         if (flags & ADDPART_FLAG_WHOLEDISK) {
 410                 err = device_create_file(pdev, &dev_attr_whole_disk);
 411                 if (err)
 412                         goto out_del;
 413         }
 414
 415         err = hd_ref_init(p);
 416         if (err) {
 417                 if (flags & ADDPART_FLAG_WHOLEDISK)
 418                         goto out_remove_file;
 419                 goto out_del;
 420         }
 421
 422         /* everything is up and running, commence */
 423         rcu_assign_pointer(ptbl->part[partno], p);
 424
 425         /* suppress uevent if the disk suppresses it */
 426         if (!dev_get_uevent_suppress(ddev))
 427                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
 428         return p;
 429
 430 out_free_info:
 431         free_part_info(p);
 432 out_free_stats:
 433         free_part_stats(p);
 434 out_free:
 435         kfree(p);
 436         return ERR_PTR(err);
 437 out_remove_file:
 438         device_remove_file(pdev, &dev_attr_whole_disk);
 439 out_del:
 440         kobject_put(p->holder_dir);
 441         device_del(pdev);
 442 out_put:
 443         put_device(pdev);
 444         return ERR_PTR(err);
 445 }
 446
 447 static bool disk_unlock_native_capacity(struct gendisk *disk)
 448 {
 449         const struct block_device_operations *bdops = disk->fops;
 450
 451         if (bdops->unlock_native_capacity &&
 452             !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
 453                 printk(KERN_CONT "enabling native capacity\n");
 454                 bdops->unlock_native_capacity(disk);
 455                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
 456                 return true;
 457         } else {
 458                 printk(KERN_CONT "truncated\n");
 459                 return false;
 460         }
 461 }
 462
 463 int blk_drop_partitions(struct gendisk *disk, struct block_device *bdev)
 464 {
 465         struct disk_part_iter piter;
 466         struct hd_struct *part;
 467         int res;
 468
 469         if (!disk_part_scan_enabled(disk))
 470                 return 0;
 471         if (bdev->bd_part_count)
 472                 return -EBUSY;
 473         res = invalidate_partition(disk, 0);
 474         if (res)
 475                 return res;
 476
 477         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 478         while ((part = disk_part_iter_next(&piter)))
 479                 delete_partition(disk, part->partno);
 480         disk_part_iter_exit(&piter);
 481
 482         return 0;
 483 }
 484
 485 static bool blk_add_partition(struct gendisk *disk, struct block_device *bdev,
 486                 struct parsed_partitions *state, int p)
 487 {
 488         sector_t size = state->parts[p].size;
 489         sector_t from = state->parts[p].from;
 490         struct hd_struct *part;
 491
 492         if (!size)
 493                 return true;
 494
 495         if (from >= get_capacity(disk)) {
 496                 printk(KERN_WARNING
 497                        "%s: p%d start %llu is beyond EOD, ",
 498                        disk->disk_name, p, (unsigned long long) from);
 499                 if (disk_unlock_native_capacity(disk))
 500                         return false;
 501                 return true;
 502         }
 503
 504         if (from + size > get_capacity(disk)) {
 505                 printk(KERN_WARNING
 506                        "%s: p%d size %llu extends beyond EOD, ",
 507                        disk->disk_name, p, (unsigned long long) size);
 508
 509                 if (disk_unlock_native_capacity(disk))
 510                         return false;
 511
 512                 /*
 513                  * We can not ignore partitions of broken tables created by for
 514                  * example camera firmware, but we limit them to the end of the
 515                  * disk to avoid creating invalid block devices.
 516                  */
 517                 size = get_capacity(disk) - from;
 518         }
 519
 520         part = add_partition(disk, p, from, size, state->parts[p].flags,
 521                              &state->parts[p].info);
 522         if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
 523                 printk(KERN_ERR " %s: p%d could not be added: %ld\n",
 524                        disk->disk_name, p, -PTR_ERR(part));
 525                 return true;
 526         }
 527
 528 #ifdef CONFIG_BLK_DEV_MD
 529         if (state->parts[p].flags & ADDPART_FLAG_RAID)
 530                 md_autodetect_dev(part_to_dev(part)->devt);
 531 #endif
 532         return true;
 533 }
 534
 535 int blk_add_partitions(struct gendisk *disk, struct block_device *bdev)
 536 {
 537         struct parsed_partitions *state;
 538         int ret = -EAGAIN, p, highest;
 539
 540         if (!disk_part_scan_enabled(disk))
 541                 return 0;
 542
 543         state = check_partition(disk, bdev);
 544         if (!state)
 545                 return 0;
 546         if (IS_ERR(state)) {
 547                 /*
 548                  * I/O error reading the partition table.  If we tried to read
 549                  * beyond EOD, retry after unlocking the native capacity.
 550                  */
 551                 if (PTR_ERR(state) == -ENOSPC) {
 552                         printk(KERN_WARNING "%s: partition table beyond EOD, ",
 553                                disk->disk_name);
 554                         if (disk_unlock_native_capacity(disk))
 555                                 return -EAGAIN;
 556                 }
 557                 return -EIO;
 558         }
 559
 560         /*
 561          * Partitions are not supported on host managed zoned block devices.
 562          */
 563         if (disk->queue->limits.zoned == BLK_ZONED_HM) {
 564                 pr_warn("%s: ignoring partition table on host managed zoned block device\n",
 565                         disk->disk_name);
 566                 ret = 0;
 567                 goto out_free_state;
 568         }
 569
 570         /*
 571          * If we read beyond EOD, try unlocking native capacity even if the
 572          * partition table was successfully read as we could be missing some
 573          * partitions.
 574          */
 575         if (state->access_beyond_eod) {
 576                 printk(KERN_WARNING
 577                        "%s: partition table partially beyond EOD, ",
 578                        disk->disk_name);
 579                 if (disk_unlock_native_capacity(disk))
 580                         goto out_free_state;
 581         }
 582
 583         /* tell userspace that the media / partition table may have changed */
 584         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 585
 586         /*
 587          * Detect the highest partition number and preallocate disk->part_tbl.
 588          * This is an optimization and not strictly necessary.
 589          */
 590         for (p = 1, highest = 0; p < state->limit; p++)
 591                 if (state->parts[p].size)
 592                         highest = p;
 593         disk_expand_part_tbl(disk, highest);
 594
 595         for (p = 1; p < state->limit; p++)
 596                 if (!blk_add_partition(disk, bdev, state, p))
 597                         goto out_free_state;
 598
 599         ret = 0;
 600 out_free_state:
 601         free_partitions(state);
 602         return ret;
 603 }
 604
 605 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
 606 {
 607         struct address_space *mapping = bdev->bd_inode->i_mapping;
 608         struct page *page;
 609
 610         page = read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_SHIFT-9)), NULL);
 611         if (!IS_ERR(page)) {
 612                 if (PageError(page))
 613                         goto fail;
 614                 p->v = page;
 615                 return (unsigned char *)page_address(page) +  ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << 9);
 616 fail:
 617                 put_page(page);
 618         }
 619         p->v = NULL;
 620         return NULL;
 621 }
 622
 623 EXPORT_SYMBOL(read_dev_sector);