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 (*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 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
90 spin_lock(&bdev->bd_size_lock);
91 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
92 bdev->bd_nr_sectors = sectors;
93 spin_unlock(&bdev->bd_size_lock);
96 static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
98 struct parsed_partitions *state;
99 int nr = DISK_MAX_PARTS;
101 state = kzalloc(sizeof(*state), GFP_KERNEL);
105 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
116 static void free_partitions(struct parsed_partitions *state)
122 static struct parsed_partitions *check_partition(struct gendisk *hd)
124 struct parsed_partitions *state;
127 state = allocate_partitions(hd);
130 state->pp_buf = (char *)__get_free_page(GFP_KERNEL);
131 if (!state->pp_buf) {
132 free_partitions(state);
135 state->pp_buf[0] = '\0';
138 snprintf(state->name, BDEVNAME_SIZE, "%s", hd->disk_name);
139 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name);
140 if (isdigit(state->name[strlen(state->name)-1]))
141 sprintf(state->name, "p");
144 while (!res && check_part[i]) {
145 memset(state->parts, 0, state->limit * sizeof(state->parts[0]));
146 res = check_part[i++](state);
149 * We have hit an I/O error which we don't report now.
150 * But record it, and let the others do their job.
158 printk(KERN_INFO "%s", state->pp_buf);
160 free_page((unsigned long)state->pp_buf);
163 if (state->access_beyond_eod)
166 * The partition is unrecognized. So report I/O errors if there were any
171 strlcat(state->pp_buf,
172 " unable to read partition table\n", PAGE_SIZE);
173 printk(KERN_INFO "%s", state->pp_buf);
176 free_page((unsigned long)state->pp_buf);
177 free_partitions(state);
181 static ssize_t part_partition_show(struct device *dev,
182 struct device_attribute *attr, char *buf)
184 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno);
187 static ssize_t part_start_show(struct device *dev,
188 struct device_attribute *attr, char *buf)
190 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect);
193 static ssize_t part_ro_show(struct device *dev,
194 struct device_attribute *attr, char *buf)
196 return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev)));
199 static ssize_t part_alignment_offset_show(struct device *dev,
200 struct device_attribute *attr, char *buf)
202 return sprintf(buf, "%u\n", bdev_alignment_offset(dev_to_bdev(dev)));
205 static ssize_t part_discard_alignment_show(struct device *dev,
206 struct device_attribute *attr, char *buf)
208 return sprintf(buf, "%u\n", bdev_discard_alignment(dev_to_bdev(dev)));
211 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
212 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
213 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
214 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
215 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
216 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
217 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
218 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
219 #ifdef CONFIG_FAIL_MAKE_REQUEST
220 static struct device_attribute dev_attr_fail =
221 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
224 static struct attribute *part_attrs[] = {
225 &dev_attr_partition.attr,
226 &dev_attr_start.attr,
229 &dev_attr_alignment_offset.attr,
230 &dev_attr_discard_alignment.attr,
232 &dev_attr_inflight.attr,
233 #ifdef CONFIG_FAIL_MAKE_REQUEST
239 static struct attribute_group part_attr_group = {
243 static const struct attribute_group *part_attr_groups[] = {
245 #ifdef CONFIG_BLK_DEV_IO_TRACE
246 &blk_trace_attr_group,
251 static void part_release(struct device *dev)
253 put_disk(dev_to_bdev(dev)->bd_disk);
254 iput(dev_to_bdev(dev)->bd_inode);
257 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
259 struct block_device *part = dev_to_bdev(dev);
261 add_uevent_var(env, "PARTN=%u", part->bd_partno);
262 if (part->bd_meta_info && part->bd_meta_info->volname[0])
263 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname);
267 struct device_type part_type = {
269 .groups = part_attr_groups,
270 .release = part_release,
271 .uevent = part_uevent,
274 static void delete_partition(struct block_device *part)
276 lockdep_assert_held(&part->bd_disk->open_mutex);
279 __invalidate_device(part, true);
281 xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
282 kobject_put(part->bd_holder_dir);
283 device_del(&part->bd_device);
286 * Remove the block device from the inode hash, so that it cannot be
287 * looked up any more even when openers still hold references.
289 remove_inode_hash(part->bd_inode);
291 put_device(&part->bd_device);
294 static ssize_t whole_disk_show(struct device *dev,
295 struct device_attribute *attr, char *buf)
299 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
302 * Must be called either with open_mutex held, before a disk can be opened or
303 * after all disk users are gone.
305 static struct block_device *add_partition(struct gendisk *disk, int partno,
306 sector_t start, sector_t len, int flags,
307 struct partition_meta_info *info)
309 dev_t devt = MKDEV(0, 0);
310 struct device *ddev = disk_to_dev(disk);
312 struct block_device *bdev;
316 lockdep_assert_held(&disk->open_mutex);
318 if (partno >= DISK_MAX_PARTS)
319 return ERR_PTR(-EINVAL);
322 * Partitions are not supported on zoned block devices that are used as
325 switch (disk->queue->limits.zoned) {
327 pr_warn("%s: partitions not supported on host managed zoned block device\n",
329 return ERR_PTR(-ENXIO);
331 pr_info("%s: disabling host aware zoned block device support due to partitions\n",
333 disk_set_zoned(disk, BLK_ZONED_NONE);
339 if (xa_load(&disk->part_tbl, partno))
340 return ERR_PTR(-EBUSY);
342 /* ensure we always have a reference to the whole disk */
343 get_device(disk_to_dev(disk));
346 bdev = bdev_alloc(disk, partno);
350 bdev->bd_start_sect = start;
351 bdev_set_nr_sectors(bdev, len);
353 pdev = &bdev->bd_device;
354 dname = dev_name(ddev);
355 if (isdigit(dname[strlen(dname) - 1]))
356 dev_set_name(pdev, "%sp%d", dname, partno);
358 dev_set_name(pdev, "%s%d", dname, partno);
360 device_initialize(pdev);
361 pdev->class = &block_class;
362 pdev->type = &part_type;
365 /* in consecutive minor range? */
366 if (bdev->bd_partno < disk->minors) {
367 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno);
369 err = blk_alloc_ext_minor();
372 devt = MKDEV(BLOCK_EXT_MAJOR, err);
378 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
379 if (!bdev->bd_meta_info)
383 /* delay uevent until 'holders' subdir is created */
384 dev_set_uevent_suppress(pdev, 1);
385 err = device_add(pdev);
390 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
391 if (!bdev->bd_holder_dir)
394 dev_set_uevent_suppress(pdev, 0);
395 if (flags & ADDPART_FLAG_WHOLEDISK) {
396 err = device_create_file(pdev, &dev_attr_whole_disk);
401 /* everything is up and running, commence */
402 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
405 bdev_add(bdev, devt);
407 /* suppress uevent if the disk suppresses it */
408 if (!dev_get_uevent_suppress(ddev))
409 kobject_uevent(&pdev->kobj, KOBJ_ADD);
413 kobject_put(bdev->bd_holder_dir);
423 static bool partition_overlaps(struct gendisk *disk, sector_t start,
424 sector_t length, int skip_partno)
426 struct block_device *part;
427 bool overlap = false;
431 xa_for_each_start(&disk->part_tbl, idx, part, 1) {
432 if (part->bd_partno != skip_partno &&
433 start < part->bd_start_sect + bdev_nr_sectors(part) &&
434 start + length > part->bd_start_sect) {
444 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
447 struct block_device *part;
450 mutex_lock(&disk->open_mutex);
451 if (!disk_live(disk)) {
456 if (partition_overlaps(disk, start, length, -1)) {
461 part = add_partition(disk, partno, start, length,
462 ADDPART_FLAG_NONE, NULL);
463 ret = PTR_ERR_OR_ZERO(part);
465 mutex_unlock(&disk->open_mutex);
469 int bdev_del_partition(struct gendisk *disk, int partno)
471 struct block_device *part = NULL;
474 mutex_lock(&disk->open_mutex);
475 part = xa_load(&disk->part_tbl, partno);
480 if (atomic_read(&part->bd_openers))
483 delete_partition(part);
486 mutex_unlock(&disk->open_mutex);
490 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
493 struct block_device *part = NULL;
496 mutex_lock(&disk->open_mutex);
497 part = xa_load(&disk->part_tbl, partno);
502 if (start != part->bd_start_sect)
506 if (partition_overlaps(disk, start, length, partno))
509 bdev_set_nr_sectors(part, length);
513 mutex_unlock(&disk->open_mutex);
517 static bool disk_unlock_native_capacity(struct gendisk *disk)
519 if (!disk->fops->unlock_native_capacity ||
520 test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) {
521 printk(KERN_CONT "truncated\n");
525 printk(KERN_CONT "enabling native capacity\n");
526 disk->fops->unlock_native_capacity(disk);
530 void blk_drop_partitions(struct gendisk *disk)
532 struct block_device *part;
535 lockdep_assert_held(&disk->open_mutex);
537 xa_for_each_start(&disk->part_tbl, idx, part, 1)
538 delete_partition(part);
541 static bool blk_add_partition(struct gendisk *disk,
542 struct parsed_partitions *state, int p)
544 sector_t size = state->parts[p].size;
545 sector_t from = state->parts[p].from;
546 struct block_device *part;
551 if (from >= get_capacity(disk)) {
553 "%s: p%d start %llu is beyond EOD, ",
554 disk->disk_name, p, (unsigned long long) from);
555 if (disk_unlock_native_capacity(disk))
560 if (from + size > get_capacity(disk)) {
562 "%s: p%d size %llu extends beyond EOD, ",
563 disk->disk_name, p, (unsigned long long) size);
565 if (disk_unlock_native_capacity(disk))
569 * We can not ignore partitions of broken tables created by for
570 * example camera firmware, but we limit them to the end of the
571 * disk to avoid creating invalid block devices.
573 size = get_capacity(disk) - from;
576 part = add_partition(disk, p, from, size, state->parts[p].flags,
577 &state->parts[p].info);
578 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
579 printk(KERN_ERR " %s: p%d could not be added: %ld\n",
580 disk->disk_name, p, -PTR_ERR(part));
584 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
585 (state->parts[p].flags & ADDPART_FLAG_RAID))
586 md_autodetect_dev(part->bd_dev);
591 static int blk_add_partitions(struct gendisk *disk)
593 struct parsed_partitions *state;
594 int ret = -EAGAIN, p;
596 if (disk->flags & GENHD_FL_NO_PART)
599 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
602 state = check_partition(disk);
607 * I/O error reading the partition table. If we tried to read
608 * beyond EOD, retry after unlocking the native capacity.
610 if (PTR_ERR(state) == -ENOSPC) {
611 printk(KERN_WARNING "%s: partition table beyond EOD, ",
613 if (disk_unlock_native_capacity(disk))
620 * Partitions are not supported on host managed zoned block devices.
622 if (disk->queue->limits.zoned == BLK_ZONED_HM) {
623 pr_warn("%s: ignoring partition table on host managed zoned block device\n",
630 * If we read beyond EOD, try unlocking native capacity even if the
631 * partition table was successfully read as we could be missing some
634 if (state->access_beyond_eod) {
636 "%s: partition table partially beyond EOD, ",
638 if (disk_unlock_native_capacity(disk))
642 /* tell userspace that the media / partition table may have changed */
643 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
645 for (p = 1; p < state->limit; p++)
646 if (!blk_add_partition(disk, state, p))
651 free_partitions(state);
655 int bdev_disk_changed(struct gendisk *disk, bool invalidate)
659 lockdep_assert_held(&disk->open_mutex);
661 if (!disk_live(disk))
665 if (disk->open_partitions)
667 sync_blockdev(disk->part0);
668 invalidate_bdev(disk->part0);
669 blk_drop_partitions(disk);
671 clear_bit(GD_NEED_PART_SCAN, &disk->state);
674 * Historically we only set the capacity to zero for devices that
675 * support partitions (independ of actually having partitions created).
676 * Doing that is rather inconsistent, but changing it broke legacy
677 * udisks polling for legacy ide-cdrom devices. Use the crude check
678 * below to get the sane behavior for most device while not breaking
679 * userspace for this particular setup.
682 if (!(disk->flags & GENHD_FL_NO_PART) ||
683 !(disk->flags & GENHD_FL_REMOVABLE))
684 set_capacity(disk, 0);
687 if (get_capacity(disk)) {
688 ret = blk_add_partitions(disk);
691 } else if (invalidate) {
693 * Tell userspace that the media / partition table may have
696 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
702 * Only exported for loop and dasd for historic reasons. Don't use in new
705 EXPORT_SYMBOL_GPL(bdev_disk_changed);
707 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
709 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping;
712 if (n >= get_capacity(state->disk)) {
713 state->access_beyond_eod = true;
717 folio = read_mapping_folio(mapping, n >> PAGE_SECTORS_SHIFT, NULL);
722 return folio_address(folio) + offset_in_folio(folio, n * SECTOR_SIZE);