1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 1991-1998 Linus Torvalds
4 * Re-organised Feb 1998 Russell King
5 * Copyright (C) 2020 Christoph Hellwig
8 #include <linux/major.h>
9 #include <linux/slab.h>
10 #include <linux/ctype.h>
11 #include <linux/vmalloc.h>
12 #include <linux/raid/detect.h>
15 static int (*const check_part[])(struct parsed_partitions *) = {
17 * Probe partition formats with tables at disk address 0
18 * that also have an ADFS boot block at 0xdc0.
20 #ifdef CONFIG_ACORN_PARTITION_ICS
23 #ifdef CONFIG_ACORN_PARTITION_POWERTEC
24 adfspart_check_POWERTEC,
26 #ifdef CONFIG_ACORN_PARTITION_EESOX
31 * Now move on to formats that only have partition info at
32 * disk address 0xdc0. Since these may also have stale
33 * PC/BIOS partition tables, they need to come before
36 #ifdef CONFIG_ACORN_PARTITION_CUMANA
37 adfspart_check_CUMANA,
39 #ifdef CONFIG_ACORN_PARTITION_ADFS
43 #ifdef CONFIG_CMDLINE_PARTITION
46 #ifdef CONFIG_EFI_PARTITION
47 efi_partition, /* this must come before msdos */
49 #ifdef CONFIG_SGI_PARTITION
52 #ifdef CONFIG_LDM_PARTITION
53 ldm_partition, /* this must come before msdos */
55 #ifdef CONFIG_MSDOS_PARTITION
58 #ifdef CONFIG_OSF_PARTITION
61 #ifdef CONFIG_SUN_PARTITION
64 #ifdef CONFIG_AMIGA_PARTITION
67 #ifdef CONFIG_ATARI_PARTITION
70 #ifdef CONFIG_MAC_PARTITION
73 #ifdef CONFIG_ULTRIX_PARTITION
76 #ifdef CONFIG_IBM_PARTITION
79 #ifdef CONFIG_KARMA_PARTITION
82 #ifdef CONFIG_SYSV68_PARTITION
88 static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
90 struct parsed_partitions *state;
91 int nr = DISK_MAX_PARTS;
93 state = kzalloc(sizeof(*state), GFP_KERNEL);
97 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
108 static void free_partitions(struct parsed_partitions *state)
114 static struct parsed_partitions *check_partition(struct gendisk *hd)
116 struct parsed_partitions *state;
119 state = allocate_partitions(hd);
122 state->pp_buf = (char *)__get_free_page(GFP_KERNEL);
123 if (!state->pp_buf) {
124 free_partitions(state);
127 state->pp_buf[0] = '\0';
130 snprintf(state->name, BDEVNAME_SIZE, "%s", hd->disk_name);
131 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name);
132 if (isdigit(state->name[strlen(state->name)-1]))
133 sprintf(state->name, "p");
136 while (!res && check_part[i]) {
137 memset(state->parts, 0, state->limit * sizeof(state->parts[0]));
138 res = check_part[i++](state);
141 * We have hit an I/O error which we don't report now.
142 * But record it, and let the others do their job.
150 printk(KERN_INFO "%s", state->pp_buf);
152 free_page((unsigned long)state->pp_buf);
155 if (state->access_beyond_eod)
158 * The partition is unrecognized. So report I/O errors if there were any
163 strlcat(state->pp_buf,
164 " unable to read partition table\n", PAGE_SIZE);
165 printk(KERN_INFO "%s", state->pp_buf);
168 free_page((unsigned long)state->pp_buf);
169 free_partitions(state);
173 static ssize_t part_partition_show(struct device *dev,
174 struct device_attribute *attr, char *buf)
176 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno);
179 static ssize_t part_start_show(struct device *dev,
180 struct device_attribute *attr, char *buf)
182 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect);
185 static ssize_t part_ro_show(struct device *dev,
186 struct device_attribute *attr, char *buf)
188 return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev)));
191 static ssize_t part_alignment_offset_show(struct device *dev,
192 struct device_attribute *attr, char *buf)
194 return sprintf(buf, "%u\n", bdev_alignment_offset(dev_to_bdev(dev)));
197 static ssize_t part_discard_alignment_show(struct device *dev,
198 struct device_attribute *attr, char *buf)
200 return sprintf(buf, "%u\n", bdev_discard_alignment(dev_to_bdev(dev)));
203 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
204 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
205 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
206 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
207 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
208 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
209 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
210 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
211 #ifdef CONFIG_FAIL_MAKE_REQUEST
212 static struct device_attribute dev_attr_fail =
213 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
216 static struct attribute *part_attrs[] = {
217 &dev_attr_partition.attr,
218 &dev_attr_start.attr,
221 &dev_attr_alignment_offset.attr,
222 &dev_attr_discard_alignment.attr,
224 &dev_attr_inflight.attr,
225 #ifdef CONFIG_FAIL_MAKE_REQUEST
231 static const struct attribute_group part_attr_group = {
235 static const struct attribute_group *part_attr_groups[] = {
237 #ifdef CONFIG_BLK_DEV_IO_TRACE
238 &blk_trace_attr_group,
243 static void part_release(struct device *dev)
245 put_disk(dev_to_bdev(dev)->bd_disk);
246 iput(dev_to_bdev(dev)->bd_inode);
249 static int part_uevent(const struct device *dev, struct kobj_uevent_env *env)
251 const struct block_device *part = dev_to_bdev(dev);
253 add_uevent_var(env, "PARTN=%u", part->bd_partno);
254 if (part->bd_meta_info && part->bd_meta_info->volname[0])
255 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname);
259 const struct device_type part_type = {
261 .groups = part_attr_groups,
262 .release = part_release,
263 .uevent = part_uevent,
266 void drop_partition(struct block_device *part)
268 lockdep_assert_held(&part->bd_disk->open_mutex);
270 xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
271 kobject_put(part->bd_holder_dir);
273 device_del(&part->bd_device);
274 put_device(&part->bd_device);
277 static ssize_t whole_disk_show(struct device *dev,
278 struct device_attribute *attr, char *buf)
282 static const DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
285 * Must be called either with open_mutex held, before a disk can be opened or
286 * after all disk users are gone.
288 static struct block_device *add_partition(struct gendisk *disk, int partno,
289 sector_t start, sector_t len, int flags,
290 struct partition_meta_info *info)
292 dev_t devt = MKDEV(0, 0);
293 struct device *ddev = disk_to_dev(disk);
295 struct block_device *bdev;
299 lockdep_assert_held(&disk->open_mutex);
301 if (partno >= DISK_MAX_PARTS)
302 return ERR_PTR(-EINVAL);
305 * Partitions are not supported on zoned block devices that are used as
308 switch (disk->queue->limits.zoned) {
310 pr_warn("%s: partitions not supported on host managed zoned block device\n",
312 return ERR_PTR(-ENXIO);
314 pr_info("%s: disabling host aware zoned block device support due to partitions\n",
316 disk_set_zoned(disk, BLK_ZONED_NONE);
322 if (xa_load(&disk->part_tbl, partno))
323 return ERR_PTR(-EBUSY);
325 /* ensure we always have a reference to the whole disk */
326 get_device(disk_to_dev(disk));
329 bdev = bdev_alloc(disk, partno);
333 bdev->bd_start_sect = start;
334 bdev_set_nr_sectors(bdev, len);
336 pdev = &bdev->bd_device;
337 dname = dev_name(ddev);
338 if (isdigit(dname[strlen(dname) - 1]))
339 dev_set_name(pdev, "%sp%d", dname, partno);
341 dev_set_name(pdev, "%s%d", dname, partno);
343 device_initialize(pdev);
344 pdev->class = &block_class;
345 pdev->type = &part_type;
348 /* in consecutive minor range? */
349 if (bdev->bd_partno < disk->minors) {
350 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno);
352 err = blk_alloc_ext_minor();
355 devt = MKDEV(BLOCK_EXT_MAJOR, err);
361 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
362 if (!bdev->bd_meta_info)
366 /* delay uevent until 'holders' subdir is created */
367 dev_set_uevent_suppress(pdev, 1);
368 err = device_add(pdev);
373 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
374 if (!bdev->bd_holder_dir)
377 dev_set_uevent_suppress(pdev, 0);
378 if (flags & ADDPART_FLAG_WHOLEDISK) {
379 err = device_create_file(pdev, &dev_attr_whole_disk);
384 /* everything is up and running, commence */
385 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
388 bdev_add(bdev, devt);
390 /* suppress uevent if the disk suppresses it */
391 if (!dev_get_uevent_suppress(ddev))
392 kobject_uevent(&pdev->kobj, KOBJ_ADD);
396 kobject_put(bdev->bd_holder_dir);
406 static bool partition_overlaps(struct gendisk *disk, sector_t start,
407 sector_t length, int skip_partno)
409 struct block_device *part;
410 bool overlap = false;
414 xa_for_each_start(&disk->part_tbl, idx, part, 1) {
415 if (part->bd_partno != skip_partno &&
416 start < part->bd_start_sect + bdev_nr_sectors(part) &&
417 start + length > part->bd_start_sect) {
427 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
430 sector_t capacity = get_capacity(disk), end;
431 struct block_device *part;
434 mutex_lock(&disk->open_mutex);
435 if (check_add_overflow(start, length, &end)) {
440 if (start >= capacity || end > capacity) {
445 if (!disk_live(disk)) {
450 if (partition_overlaps(disk, start, length, -1)) {
455 part = add_partition(disk, partno, start, length,
456 ADDPART_FLAG_NONE, NULL);
457 ret = PTR_ERR_OR_ZERO(part);
459 mutex_unlock(&disk->open_mutex);
463 int bdev_del_partition(struct gendisk *disk, int partno)
465 struct block_device *part = NULL;
468 mutex_lock(&disk->open_mutex);
469 part = xa_load(&disk->part_tbl, partno);
474 if (atomic_read(&part->bd_openers))
478 * We verified that @part->bd_openers is zero above and so
479 * @part->bd_holder{_ops} can't be set. And since we hold
480 * @disk->open_mutex the device can't be claimed by anyone.
482 * So no need to call @part->bd_holder_ops->mark_dead() here.
483 * Just delete the partition and invalidate it.
486 remove_inode_hash(part->bd_inode);
487 invalidate_bdev(part);
488 drop_partition(part);
491 mutex_unlock(&disk->open_mutex);
495 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
498 struct block_device *part = NULL;
501 mutex_lock(&disk->open_mutex);
502 part = xa_load(&disk->part_tbl, partno);
507 if (start != part->bd_start_sect)
511 if (partition_overlaps(disk, start, length, partno))
514 bdev_set_nr_sectors(part, length);
518 mutex_unlock(&disk->open_mutex);
522 static bool disk_unlock_native_capacity(struct gendisk *disk)
524 if (!disk->fops->unlock_native_capacity ||
525 test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) {
526 printk(KERN_CONT "truncated\n");
530 printk(KERN_CONT "enabling native capacity\n");
531 disk->fops->unlock_native_capacity(disk);
535 static bool blk_add_partition(struct gendisk *disk,
536 struct parsed_partitions *state, int p)
538 sector_t size = state->parts[p].size;
539 sector_t from = state->parts[p].from;
540 struct block_device *part;
545 if (from >= get_capacity(disk)) {
547 "%s: p%d start %llu is beyond EOD, ",
548 disk->disk_name, p, (unsigned long long) from);
549 if (disk_unlock_native_capacity(disk))
554 if (from + size > get_capacity(disk)) {
556 "%s: p%d size %llu extends beyond EOD, ",
557 disk->disk_name, p, (unsigned long long) size);
559 if (disk_unlock_native_capacity(disk))
563 * We can not ignore partitions of broken tables created by for
564 * example camera firmware, but we limit them to the end of the
565 * disk to avoid creating invalid block devices.
567 size = get_capacity(disk) - from;
570 part = add_partition(disk, p, from, size, state->parts[p].flags,
571 &state->parts[p].info);
572 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
573 printk(KERN_ERR " %s: p%d could not be added: %ld\n",
574 disk->disk_name, p, -PTR_ERR(part));
578 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
579 (state->parts[p].flags & ADDPART_FLAG_RAID))
580 md_autodetect_dev(part->bd_dev);
585 static int blk_add_partitions(struct gendisk *disk)
587 struct parsed_partitions *state;
588 int ret = -EAGAIN, p;
590 if (disk->flags & GENHD_FL_NO_PART)
593 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
596 state = check_partition(disk);
601 * I/O error reading the partition table. If we tried to read
602 * beyond EOD, retry after unlocking the native capacity.
604 if (PTR_ERR(state) == -ENOSPC) {
605 printk(KERN_WARNING "%s: partition table beyond EOD, ",
607 if (disk_unlock_native_capacity(disk))
614 * Partitions are not supported on host managed zoned block devices.
616 if (disk->queue->limits.zoned == BLK_ZONED_HM) {
617 pr_warn("%s: ignoring partition table on host managed zoned block device\n",
624 * If we read beyond EOD, try unlocking native capacity even if the
625 * partition table was successfully read as we could be missing some
628 if (state->access_beyond_eod) {
630 "%s: partition table partially beyond EOD, ",
632 if (disk_unlock_native_capacity(disk))
636 /* tell userspace that the media / partition table may have changed */
637 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
639 for (p = 1; p < state->limit; p++)
640 if (!blk_add_partition(disk, state, p))
645 free_partitions(state);
649 int bdev_disk_changed(struct gendisk *disk, bool invalidate)
651 struct block_device *part;
655 lockdep_assert_held(&disk->open_mutex);
657 if (!disk_live(disk))
661 if (disk->open_partitions)
663 sync_blockdev(disk->part0);
664 invalidate_bdev(disk->part0);
666 xa_for_each_start(&disk->part_tbl, idx, part, 1) {
668 * Remove the block device from the inode hash, so that
669 * it cannot be looked up any more even when openers
670 * still hold references.
672 remove_inode_hash(part->bd_inode);
675 * If @disk->open_partitions isn't elevated but there's
676 * still an active holder of that block device things
679 WARN_ON_ONCE(atomic_read(&part->bd_openers));
680 invalidate_bdev(part);
681 drop_partition(part);
683 clear_bit(GD_NEED_PART_SCAN, &disk->state);
686 * Historically we only set the capacity to zero for devices that
687 * support partitions (independ of actually having partitions created).
688 * Doing that is rather inconsistent, but changing it broke legacy
689 * udisks polling for legacy ide-cdrom devices. Use the crude check
690 * below to get the sane behavior for most device while not breaking
691 * userspace for this particular setup.
694 if (!(disk->flags & GENHD_FL_NO_PART) ||
695 !(disk->flags & GENHD_FL_REMOVABLE))
696 set_capacity(disk, 0);
699 if (get_capacity(disk)) {
700 ret = blk_add_partitions(disk);
703 } else if (invalidate) {
705 * Tell userspace that the media / partition table may have
708 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
714 * Only exported for loop and dasd for historic reasons. Don't use in new
717 EXPORT_SYMBOL_GPL(bdev_disk_changed);
719 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
721 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping;
724 if (n >= get_capacity(state->disk)) {
725 state->access_beyond_eod = true;
729 folio = read_mapping_folio(mapping, n >> PAGE_SECTORS_SHIFT, NULL);
734 return folio_address(folio) + offset_in_folio(folio, n * SECTOR_SIZE);