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/slab.h>
9 #include <linux/ctype.h>
10 #include <linux/genhd.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 spin_unlock(&bdev->bd_size_lock);
96 static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
98 struct parsed_partitions *state;
101 state = kzalloc(sizeof(*state), GFP_KERNEL);
105 nr = disk_max_parts(hd);
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 struct block_device *bdev = dev_to_bdev(dev);
205 return sprintf(buf, "%u\n",
206 queue_limit_alignment_offset(&bdev->bd_disk->queue->limits,
207 bdev->bd_start_sect));
210 static ssize_t part_discard_alignment_show(struct device *dev,
211 struct device_attribute *attr, char *buf)
213 struct block_device *bdev = dev_to_bdev(dev);
215 return sprintf(buf, "%u\n",
216 queue_limit_discard_alignment(&bdev->bd_disk->queue->limits,
217 bdev->bd_start_sect));
220 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
221 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
222 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
223 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
224 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
225 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
226 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
227 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
228 #ifdef CONFIG_FAIL_MAKE_REQUEST
229 static struct device_attribute dev_attr_fail =
230 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
233 static struct attribute *part_attrs[] = {
234 &dev_attr_partition.attr,
235 &dev_attr_start.attr,
238 &dev_attr_alignment_offset.attr,
239 &dev_attr_discard_alignment.attr,
241 &dev_attr_inflight.attr,
242 #ifdef CONFIG_FAIL_MAKE_REQUEST
248 static struct attribute_group part_attr_group = {
252 static const struct attribute_group *part_attr_groups[] = {
254 #ifdef CONFIG_BLK_DEV_IO_TRACE
255 &blk_trace_attr_group,
260 static void part_release(struct device *dev)
262 put_disk(dev_to_bdev(dev)->bd_disk);
263 iput(dev_to_bdev(dev)->bd_inode);
266 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
268 struct block_device *part = dev_to_bdev(dev);
270 add_uevent_var(env, "PARTN=%u", part->bd_partno);
271 if (part->bd_meta_info && part->bd_meta_info->volname[0])
272 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname);
276 struct device_type part_type = {
278 .groups = part_attr_groups,
279 .release = part_release,
280 .uevent = part_uevent,
283 static void delete_partition(struct block_device *part)
285 lockdep_assert_held(&part->bd_disk->open_mutex);
288 __invalidate_device(part, true);
290 xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
291 kobject_put(part->bd_holder_dir);
292 device_del(&part->bd_device);
295 * Remove the block device from the inode hash, so that it cannot be
296 * looked up any more even when openers still hold references.
298 remove_inode_hash(part->bd_inode);
300 put_device(&part->bd_device);
303 static ssize_t whole_disk_show(struct device *dev,
304 struct device_attribute *attr, char *buf)
308 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
311 * Must be called either with open_mutex held, before a disk can be opened or
312 * after all disk users are gone.
314 static struct block_device *add_partition(struct gendisk *disk, int partno,
315 sector_t start, sector_t len, int flags,
316 struct partition_meta_info *info)
318 dev_t devt = MKDEV(0, 0);
319 struct device *ddev = disk_to_dev(disk);
321 struct block_device *bdev;
325 lockdep_assert_held(&disk->open_mutex);
327 if (partno >= disk_max_parts(disk))
328 return ERR_PTR(-EINVAL);
331 * Partitions are not supported on zoned block devices that are used as
334 switch (disk->queue->limits.zoned) {
336 pr_warn("%s: partitions not supported on host managed zoned block device\n",
338 return ERR_PTR(-ENXIO);
340 pr_info("%s: disabling host aware zoned block device support due to partitions\n",
342 blk_queue_set_zoned(disk, BLK_ZONED_NONE);
348 if (xa_load(&disk->part_tbl, partno))
349 return ERR_PTR(-EBUSY);
351 /* ensure we always have a reference to the whole disk */
352 get_device(disk_to_dev(disk));
355 bdev = bdev_alloc(disk, partno);
359 bdev->bd_start_sect = start;
360 bdev_set_nr_sectors(bdev, len);
362 pdev = &bdev->bd_device;
363 dname = dev_name(ddev);
364 if (isdigit(dname[strlen(dname) - 1]))
365 dev_set_name(pdev, "%sp%d", dname, partno);
367 dev_set_name(pdev, "%s%d", dname, partno);
369 device_initialize(pdev);
370 pdev->class = &block_class;
371 pdev->type = &part_type;
374 /* in consecutive minor range? */
375 if (bdev->bd_partno < disk->minors) {
376 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno);
378 err = blk_alloc_ext_minor();
381 devt = MKDEV(BLOCK_EXT_MAJOR, err);
387 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
388 if (!bdev->bd_meta_info)
392 /* delay uevent until 'holders' subdir is created */
393 dev_set_uevent_suppress(pdev, 1);
394 err = device_add(pdev);
399 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
400 if (!bdev->bd_holder_dir)
403 dev_set_uevent_suppress(pdev, 0);
404 if (flags & ADDPART_FLAG_WHOLEDISK) {
405 err = device_create_file(pdev, &dev_attr_whole_disk);
410 /* everything is up and running, commence */
411 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
414 bdev_add(bdev, devt);
416 /* suppress uevent if the disk suppresses it */
417 if (!dev_get_uevent_suppress(ddev))
418 kobject_uevent(&pdev->kobj, KOBJ_ADD);
422 kobject_put(bdev->bd_holder_dir);
431 static bool partition_overlaps(struct gendisk *disk, sector_t start,
432 sector_t length, int skip_partno)
434 struct block_device *part;
435 bool overlap = false;
439 xa_for_each_start(&disk->part_tbl, idx, part, 1) {
440 if (part->bd_partno != skip_partno &&
441 start < part->bd_start_sect + bdev_nr_sectors(part) &&
442 start + length > part->bd_start_sect) {
452 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
455 struct block_device *part;
458 mutex_lock(&disk->open_mutex);
459 if (!disk_live(disk)) {
464 if (partition_overlaps(disk, start, length, -1)) {
469 part = add_partition(disk, partno, start, length,
470 ADDPART_FLAG_NONE, NULL);
471 ret = PTR_ERR_OR_ZERO(part);
473 mutex_unlock(&disk->open_mutex);
477 int bdev_del_partition(struct gendisk *disk, int partno)
479 struct block_device *part = NULL;
482 mutex_lock(&disk->open_mutex);
483 part = xa_load(&disk->part_tbl, partno);
488 if (part->bd_openers)
491 delete_partition(part);
494 mutex_unlock(&disk->open_mutex);
498 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
501 struct block_device *part = NULL;
504 mutex_lock(&disk->open_mutex);
505 part = xa_load(&disk->part_tbl, partno);
510 if (start != part->bd_start_sect)
514 if (partition_overlaps(disk, start, length, partno))
517 bdev_set_nr_sectors(part, length);
521 mutex_unlock(&disk->open_mutex);
525 static bool disk_unlock_native_capacity(struct gendisk *disk)
527 const struct block_device_operations *bdops = disk->fops;
529 if (bdops->unlock_native_capacity &&
530 !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
531 printk(KERN_CONT "enabling native capacity\n");
532 bdops->unlock_native_capacity(disk);
533 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
536 printk(KERN_CONT "truncated\n");
541 void blk_drop_partitions(struct gendisk *disk)
543 struct block_device *part;
546 lockdep_assert_held(&disk->open_mutex);
548 xa_for_each_start(&disk->part_tbl, idx, part, 1)
549 delete_partition(part);
552 static bool blk_add_partition(struct gendisk *disk,
553 struct parsed_partitions *state, int p)
555 sector_t size = state->parts[p].size;
556 sector_t from = state->parts[p].from;
557 struct block_device *part;
562 if (from >= get_capacity(disk)) {
564 "%s: p%d start %llu is beyond EOD, ",
565 disk->disk_name, p, (unsigned long long) from);
566 if (disk_unlock_native_capacity(disk))
571 if (from + size > get_capacity(disk)) {
573 "%s: p%d size %llu extends beyond EOD, ",
574 disk->disk_name, p, (unsigned long long) size);
576 if (disk_unlock_native_capacity(disk))
580 * We can not ignore partitions of broken tables created by for
581 * example camera firmware, but we limit them to the end of the
582 * disk to avoid creating invalid block devices.
584 size = get_capacity(disk) - from;
587 part = add_partition(disk, p, from, size, state->parts[p].flags,
588 &state->parts[p].info);
589 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
590 printk(KERN_ERR " %s: p%d could not be added: %ld\n",
591 disk->disk_name, p, -PTR_ERR(part));
595 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
596 (state->parts[p].flags & ADDPART_FLAG_RAID))
597 md_autodetect_dev(part->bd_dev);
602 static int blk_add_partitions(struct gendisk *disk)
604 struct parsed_partitions *state;
605 int ret = -EAGAIN, p;
607 if (!disk_part_scan_enabled(disk))
610 state = check_partition(disk);
615 * I/O error reading the partition table. If we tried to read
616 * beyond EOD, retry after unlocking the native capacity.
618 if (PTR_ERR(state) == -ENOSPC) {
619 printk(KERN_WARNING "%s: partition table beyond EOD, ",
621 if (disk_unlock_native_capacity(disk))
628 * Partitions are not supported on host managed zoned block devices.
630 if (disk->queue->limits.zoned == BLK_ZONED_HM) {
631 pr_warn("%s: ignoring partition table on host managed zoned block device\n",
638 * If we read beyond EOD, try unlocking native capacity even if the
639 * partition table was successfully read as we could be missing some
642 if (state->access_beyond_eod) {
644 "%s: partition table partially beyond EOD, ",
646 if (disk_unlock_native_capacity(disk))
650 /* tell userspace that the media / partition table may have changed */
651 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
653 for (p = 1; p < state->limit; p++)
654 if (!blk_add_partition(disk, state, p))
659 free_partitions(state);
663 int bdev_disk_changed(struct gendisk *disk, bool invalidate)
667 lockdep_assert_held(&disk->open_mutex);
669 if (!disk_live(disk))
673 if (disk->open_partitions)
675 sync_blockdev(disk->part0);
676 invalidate_bdev(disk->part0);
677 blk_drop_partitions(disk);
679 clear_bit(GD_NEED_PART_SCAN, &disk->state);
682 * Historically we only set the capacity to zero for devices that
683 * support partitions (independ of actually having partitions created).
684 * Doing that is rather inconsistent, but changing it broke legacy
685 * udisks polling for legacy ide-cdrom devices. Use the crude check
686 * below to get the sane behavior for most device while not breaking
687 * userspace for this particular setup.
690 if (disk_part_scan_enabled(disk) ||
691 !(disk->flags & GENHD_FL_REMOVABLE))
692 set_capacity(disk, 0);
695 if (get_capacity(disk)) {
696 ret = blk_add_partitions(disk);
699 } else if (invalidate) {
701 * Tell userspace that the media / partition table may have
704 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
710 * Only exported for loop and dasd for historic reasons. Don't use in new
713 EXPORT_SYMBOL_GPL(bdev_disk_changed);
715 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
717 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping;
720 if (n >= get_capacity(state->disk)) {
721 state->access_beyond_eod = true;
725 page = read_mapping_page(mapping,
726 (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL);
733 return (unsigned char *)page_address(page) +
734 ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT);