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/blktrace_api.h>
13 #include <linux/raid/detect.h>
16 static int (*check_part[])(struct parsed_partitions *) = {
18 * Probe partition formats with tables at disk address 0
19 * that also have an ADFS boot block at 0xdc0.
21 #ifdef CONFIG_ACORN_PARTITION_ICS
24 #ifdef CONFIG_ACORN_PARTITION_POWERTEC
25 adfspart_check_POWERTEC,
27 #ifdef CONFIG_ACORN_PARTITION_EESOX
32 * Now move on to formats that only have partition info at
33 * disk address 0xdc0. Since these may also have stale
34 * PC/BIOS partition tables, they need to come before
37 #ifdef CONFIG_ACORN_PARTITION_CUMANA
38 adfspart_check_CUMANA,
40 #ifdef CONFIG_ACORN_PARTITION_ADFS
44 #ifdef CONFIG_CMDLINE_PARTITION
47 #ifdef CONFIG_EFI_PARTITION
48 efi_partition, /* this must come before msdos */
50 #ifdef CONFIG_SGI_PARTITION
53 #ifdef CONFIG_LDM_PARTITION
54 ldm_partition, /* this must come before msdos */
56 #ifdef CONFIG_MSDOS_PARTITION
59 #ifdef CONFIG_OSF_PARTITION
62 #ifdef CONFIG_SUN_PARTITION
65 #ifdef CONFIG_AMIGA_PARTITION
68 #ifdef CONFIG_ATARI_PARTITION
71 #ifdef CONFIG_MAC_PARTITION
74 #ifdef CONFIG_ULTRIX_PARTITION
77 #ifdef CONFIG_IBM_PARTITION
80 #ifdef CONFIG_KARMA_PARTITION
83 #ifdef CONFIG_SYSV68_PARTITION
89 static void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
91 spin_lock(&bdev->bd_size_lock);
92 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
93 bdev->bd_nr_sectors = sectors;
94 spin_unlock(&bdev->bd_size_lock);
97 static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
99 struct parsed_partitions *state;
100 int nr = DISK_MAX_PARTS;
102 state = kzalloc(sizeof(*state), GFP_KERNEL);
106 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
117 static void free_partitions(struct parsed_partitions *state)
123 static struct parsed_partitions *check_partition(struct gendisk *hd)
125 struct parsed_partitions *state;
128 state = allocate_partitions(hd);
131 state->pp_buf = (char *)__get_free_page(GFP_KERNEL);
132 if (!state->pp_buf) {
133 free_partitions(state);
136 state->pp_buf[0] = '\0';
139 snprintf(state->name, BDEVNAME_SIZE, "%s", hd->disk_name);
140 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name);
141 if (isdigit(state->name[strlen(state->name)-1]))
142 sprintf(state->name, "p");
145 while (!res && check_part[i]) {
146 memset(state->parts, 0, state->limit * sizeof(state->parts[0]));
147 res = check_part[i++](state);
150 * We have hit an I/O error which we don't report now.
151 * But record it, and let the others do their job.
159 printk(KERN_INFO "%s", state->pp_buf);
161 free_page((unsigned long)state->pp_buf);
164 if (state->access_beyond_eod)
167 * The partition is unrecognized. So report I/O errors if there were any
172 strlcat(state->pp_buf,
173 " unable to read partition table\n", PAGE_SIZE);
174 printk(KERN_INFO "%s", state->pp_buf);
177 free_page((unsigned long)state->pp_buf);
178 free_partitions(state);
182 static ssize_t part_partition_show(struct device *dev,
183 struct device_attribute *attr, char *buf)
185 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno);
188 static ssize_t part_start_show(struct device *dev,
189 struct device_attribute *attr, char *buf)
191 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect);
194 static ssize_t part_ro_show(struct device *dev,
195 struct device_attribute *attr, char *buf)
197 return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev)));
200 static ssize_t part_alignment_offset_show(struct device *dev,
201 struct device_attribute *attr, char *buf)
203 return sprintf(buf, "%u\n", bdev_alignment_offset(dev_to_bdev(dev)));
206 static ssize_t part_discard_alignment_show(struct device *dev,
207 struct device_attribute *attr, char *buf)
209 return sprintf(buf, "%u\n", bdev_discard_alignment(dev_to_bdev(dev)));
212 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
213 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
214 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
215 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
216 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
217 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
218 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
219 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
220 #ifdef CONFIG_FAIL_MAKE_REQUEST
221 static struct device_attribute dev_attr_fail =
222 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
225 static struct attribute *part_attrs[] = {
226 &dev_attr_partition.attr,
227 &dev_attr_start.attr,
230 &dev_attr_alignment_offset.attr,
231 &dev_attr_discard_alignment.attr,
233 &dev_attr_inflight.attr,
234 #ifdef CONFIG_FAIL_MAKE_REQUEST
240 static struct attribute_group part_attr_group = {
244 static const struct attribute_group *part_attr_groups[] = {
246 #ifdef CONFIG_BLK_DEV_IO_TRACE
247 &blk_trace_attr_group,
252 static void part_release(struct device *dev)
254 put_disk(dev_to_bdev(dev)->bd_disk);
255 iput(dev_to_bdev(dev)->bd_inode);
258 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
260 struct block_device *part = dev_to_bdev(dev);
262 add_uevent_var(env, "PARTN=%u", part->bd_partno);
263 if (part->bd_meta_info && part->bd_meta_info->volname[0])
264 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname);
268 struct device_type part_type = {
270 .groups = part_attr_groups,
271 .release = part_release,
272 .uevent = part_uevent,
275 static void delete_partition(struct block_device *part)
277 lockdep_assert_held(&part->bd_disk->open_mutex);
280 __invalidate_device(part, true);
282 xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
283 kobject_put(part->bd_holder_dir);
284 device_del(&part->bd_device);
287 * Remove the block device from the inode hash, so that it cannot be
288 * looked up any more even when openers still hold references.
290 remove_inode_hash(part->bd_inode);
292 put_device(&part->bd_device);
295 static ssize_t whole_disk_show(struct device *dev,
296 struct device_attribute *attr, char *buf)
300 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
303 * Must be called either with open_mutex held, before a disk can be opened or
304 * after all disk users are gone.
306 static struct block_device *add_partition(struct gendisk *disk, int partno,
307 sector_t start, sector_t len, int flags,
308 struct partition_meta_info *info)
310 dev_t devt = MKDEV(0, 0);
311 struct device *ddev = disk_to_dev(disk);
313 struct block_device *bdev;
317 lockdep_assert_held(&disk->open_mutex);
319 if (partno >= DISK_MAX_PARTS)
320 return ERR_PTR(-EINVAL);
323 * Partitions are not supported on zoned block devices that are used as
326 switch (disk->queue->limits.zoned) {
328 pr_warn("%s: partitions not supported on host managed zoned block device\n",
330 return ERR_PTR(-ENXIO);
332 pr_info("%s: disabling host aware zoned block device support due to partitions\n",
334 blk_queue_set_zoned(disk, BLK_ZONED_NONE);
340 if (xa_load(&disk->part_tbl, partno))
341 return ERR_PTR(-EBUSY);
343 /* ensure we always have a reference to the whole disk */
344 get_device(disk_to_dev(disk));
347 bdev = bdev_alloc(disk, partno);
351 bdev->bd_start_sect = start;
352 bdev_set_nr_sectors(bdev, len);
354 pdev = &bdev->bd_device;
355 dname = dev_name(ddev);
356 if (isdigit(dname[strlen(dname) - 1]))
357 dev_set_name(pdev, "%sp%d", dname, partno);
359 dev_set_name(pdev, "%s%d", dname, partno);
361 device_initialize(pdev);
362 pdev->class = &block_class;
363 pdev->type = &part_type;
366 /* in consecutive minor range? */
367 if (bdev->bd_partno < disk->minors) {
368 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno);
370 err = blk_alloc_ext_minor();
373 devt = MKDEV(BLOCK_EXT_MAJOR, err);
379 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
380 if (!bdev->bd_meta_info)
384 /* delay uevent until 'holders' subdir is created */
385 dev_set_uevent_suppress(pdev, 1);
386 err = device_add(pdev);
391 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
392 if (!bdev->bd_holder_dir)
395 dev_set_uevent_suppress(pdev, 0);
396 if (flags & ADDPART_FLAG_WHOLEDISK) {
397 err = device_create_file(pdev, &dev_attr_whole_disk);
402 /* everything is up and running, commence */
403 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
406 bdev_add(bdev, devt);
408 /* suppress uevent if the disk suppresses it */
409 if (!dev_get_uevent_suppress(ddev))
410 kobject_uevent(&pdev->kobj, KOBJ_ADD);
414 kobject_put(bdev->bd_holder_dir);
424 static bool partition_overlaps(struct gendisk *disk, sector_t start,
425 sector_t length, int skip_partno)
427 struct block_device *part;
428 bool overlap = false;
432 xa_for_each_start(&disk->part_tbl, idx, part, 1) {
433 if (part->bd_partno != skip_partno &&
434 start < part->bd_start_sect + bdev_nr_sectors(part) &&
435 start + length > part->bd_start_sect) {
445 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
448 struct block_device *part;
451 mutex_lock(&disk->open_mutex);
452 if (!disk_live(disk)) {
457 if (partition_overlaps(disk, start, length, -1)) {
462 part = add_partition(disk, partno, start, length,
463 ADDPART_FLAG_NONE, NULL);
464 ret = PTR_ERR_OR_ZERO(part);
466 mutex_unlock(&disk->open_mutex);
470 int bdev_del_partition(struct gendisk *disk, int partno)
472 struct block_device *part = NULL;
475 mutex_lock(&disk->open_mutex);
476 part = xa_load(&disk->part_tbl, partno);
481 if (atomic_read(&part->bd_openers))
484 delete_partition(part);
487 mutex_unlock(&disk->open_mutex);
491 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
494 struct block_device *part = NULL;
497 mutex_lock(&disk->open_mutex);
498 part = xa_load(&disk->part_tbl, partno);
503 if (start != part->bd_start_sect)
507 if (partition_overlaps(disk, start, length, partno))
510 bdev_set_nr_sectors(part, length);
514 mutex_unlock(&disk->open_mutex);
518 static bool disk_unlock_native_capacity(struct gendisk *disk)
520 if (!disk->fops->unlock_native_capacity ||
521 test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) {
522 printk(KERN_CONT "truncated\n");
526 printk(KERN_CONT "enabling native capacity\n");
527 disk->fops->unlock_native_capacity(disk);
531 void blk_drop_partitions(struct gendisk *disk)
533 struct block_device *part;
536 lockdep_assert_held(&disk->open_mutex);
538 xa_for_each_start(&disk->part_tbl, idx, part, 1)
539 delete_partition(part);
542 static bool blk_add_partition(struct gendisk *disk,
543 struct parsed_partitions *state, int p)
545 sector_t size = state->parts[p].size;
546 sector_t from = state->parts[p].from;
547 struct block_device *part;
552 if (from >= get_capacity(disk)) {
554 "%s: p%d start %llu is beyond EOD, ",
555 disk->disk_name, p, (unsigned long long) from);
556 if (disk_unlock_native_capacity(disk))
561 if (from + size > get_capacity(disk)) {
563 "%s: p%d size %llu extends beyond EOD, ",
564 disk->disk_name, p, (unsigned long long) size);
566 if (disk_unlock_native_capacity(disk))
570 * We can not ignore partitions of broken tables created by for
571 * example camera firmware, but we limit them to the end of the
572 * disk to avoid creating invalid block devices.
574 size = get_capacity(disk) - from;
577 part = add_partition(disk, p, from, size, state->parts[p].flags,
578 &state->parts[p].info);
579 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
580 printk(KERN_ERR " %s: p%d could not be added: %ld\n",
581 disk->disk_name, p, -PTR_ERR(part));
585 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
586 (state->parts[p].flags & ADDPART_FLAG_RAID))
587 md_autodetect_dev(part->bd_dev);
592 static int blk_add_partitions(struct gendisk *disk)
594 struct parsed_partitions *state;
595 int ret = -EAGAIN, p;
597 if (disk->flags & GENHD_FL_NO_PART)
600 state = check_partition(disk);
605 * I/O error reading the partition table. If we tried to read
606 * beyond EOD, retry after unlocking the native capacity.
608 if (PTR_ERR(state) == -ENOSPC) {
609 printk(KERN_WARNING "%s: partition table beyond EOD, ",
611 if (disk_unlock_native_capacity(disk))
618 * Partitions are not supported on host managed zoned block devices.
620 if (disk->queue->limits.zoned == BLK_ZONED_HM) {
621 pr_warn("%s: ignoring partition table on host managed zoned block device\n",
628 * If we read beyond EOD, try unlocking native capacity even if the
629 * partition table was successfully read as we could be missing some
632 if (state->access_beyond_eod) {
634 "%s: partition table partially beyond EOD, ",
636 if (disk_unlock_native_capacity(disk))
640 /* tell userspace that the media / partition table may have changed */
641 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
643 for (p = 1; p < state->limit; p++)
644 if (!blk_add_partition(disk, state, p))
649 free_partitions(state);
653 int bdev_disk_changed(struct gendisk *disk, bool invalidate)
657 lockdep_assert_held(&disk->open_mutex);
659 if (!disk_live(disk))
663 if (disk->open_partitions)
665 sync_blockdev(disk->part0);
666 invalidate_bdev(disk->part0);
667 blk_drop_partitions(disk);
669 clear_bit(GD_NEED_PART_SCAN, &disk->state);
672 * Historically we only set the capacity to zero for devices that
673 * support partitions (independ of actually having partitions created).
674 * Doing that is rather inconsistent, but changing it broke legacy
675 * udisks polling for legacy ide-cdrom devices. Use the crude check
676 * below to get the sane behavior for most device while not breaking
677 * userspace for this particular setup.
680 if (!(disk->flags & GENHD_FL_NO_PART) ||
681 !(disk->flags & GENHD_FL_REMOVABLE))
682 set_capacity(disk, 0);
685 if (get_capacity(disk)) {
686 ret = blk_add_partitions(disk);
689 } else if (invalidate) {
691 * Tell userspace that the media / partition table may have
694 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
700 * Only exported for loop and dasd for historic reasons. Don't use in new
703 EXPORT_SYMBOL_GPL(bdev_disk_changed);
705 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
707 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping;
710 if (n >= get_capacity(state->disk)) {
711 state->access_beyond_eod = true;
715 page = read_mapping_page(mapping,
716 (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL);
723 return (unsigned char *)page_address(page) +
724 ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT);