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)
93 spin_lock_irqsave(&bdev->bd_size_lock, flags);
94 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
95 spin_unlock_irqrestore(&bdev->bd_size_lock, flags);
98 static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
100 struct parsed_partitions *state;
103 state = kzalloc(sizeof(*state), GFP_KERNEL);
107 nr = disk_max_parts(hd);
108 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
119 static void free_partitions(struct parsed_partitions *state)
125 static struct parsed_partitions *check_partition(struct gendisk *hd,
126 struct block_device *bdev)
128 struct parsed_partitions *state;
131 state = allocate_partitions(hd);
134 state->pp_buf = (char *)__get_free_page(GFP_KERNEL);
135 if (!state->pp_buf) {
136 free_partitions(state);
139 state->pp_buf[0] = '\0';
142 disk_name(hd, 0, state->name);
143 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name);
144 if (isdigit(state->name[strlen(state->name)-1]))
145 sprintf(state->name, "p");
148 while (!res && check_part[i]) {
149 memset(state->parts, 0, state->limit * sizeof(state->parts[0]));
150 res = check_part[i++](state);
153 * We have hit an I/O error which we don't report now.
154 * But record it, and let the others do their job.
162 printk(KERN_INFO "%s", state->pp_buf);
164 free_page((unsigned long)state->pp_buf);
167 if (state->access_beyond_eod)
170 * The partition is unrecognized. So report I/O errors if there were any
175 strlcat(state->pp_buf,
176 " unable to read partition table\n", PAGE_SIZE);
177 printk(KERN_INFO "%s", state->pp_buf);
180 free_page((unsigned long)state->pp_buf);
181 free_partitions(state);
185 static ssize_t part_partition_show(struct device *dev,
186 struct device_attribute *attr, char *buf)
188 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno);
191 static ssize_t part_start_show(struct device *dev,
192 struct device_attribute *attr, char *buf)
194 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect);
197 static ssize_t part_ro_show(struct device *dev,
198 struct device_attribute *attr, char *buf)
200 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_read_only);
203 static ssize_t part_alignment_offset_show(struct device *dev,
204 struct device_attribute *attr, char *buf)
206 struct block_device *bdev = dev_to_bdev(dev);
208 return sprintf(buf, "%u\n",
209 queue_limit_alignment_offset(&bdev->bd_disk->queue->limits,
210 bdev->bd_start_sect));
213 static ssize_t part_discard_alignment_show(struct device *dev,
214 struct device_attribute *attr, char *buf)
216 struct block_device *bdev = dev_to_bdev(dev);
218 return sprintf(buf, "%u\n",
219 queue_limit_discard_alignment(&bdev->bd_disk->queue->limits,
220 bdev->bd_start_sect));
223 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
224 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
225 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
226 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
227 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
228 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
229 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
230 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
231 #ifdef CONFIG_FAIL_MAKE_REQUEST
232 static struct device_attribute dev_attr_fail =
233 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
236 static struct attribute *part_attrs[] = {
237 &dev_attr_partition.attr,
238 &dev_attr_start.attr,
241 &dev_attr_alignment_offset.attr,
242 &dev_attr_discard_alignment.attr,
244 &dev_attr_inflight.attr,
245 #ifdef CONFIG_FAIL_MAKE_REQUEST
251 static struct attribute_group part_attr_group = {
255 static const struct attribute_group *part_attr_groups[] = {
257 #ifdef CONFIG_BLK_DEV_IO_TRACE
258 &blk_trace_attr_group,
263 static void part_release(struct device *dev)
265 blk_free_devt(dev->devt);
266 bdput(dev_to_bdev(dev));
269 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
271 struct block_device *part = dev_to_bdev(dev);
273 add_uevent_var(env, "PARTN=%u", part->bd_partno);
274 if (part->bd_meta_info && part->bd_meta_info->volname[0])
275 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname);
279 struct device_type part_type = {
281 .groups = part_attr_groups,
282 .release = part_release,
283 .uevent = part_uevent,
287 * Must be called either with bd_mutex held, before a disk can be opened or
288 * after all disk users are gone.
290 void delete_partition(struct block_device *part)
292 struct gendisk *disk = part->bd_disk;
293 struct disk_part_tbl *ptbl =
294 rcu_dereference_protected(disk->part_tbl, 1);
296 rcu_assign_pointer(ptbl->part[part->bd_partno], NULL);
297 rcu_assign_pointer(ptbl->last_lookup, NULL);
299 kobject_put(part->bd_holder_dir);
300 device_del(&part->bd_device);
303 * Remove the block device from the inode hash, so that it cannot be
304 * looked up any more even when openers still hold references.
306 remove_inode_hash(part->bd_inode);
308 put_device(&part->bd_device);
311 static ssize_t whole_disk_show(struct device *dev,
312 struct device_attribute *attr, char *buf)
316 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
319 * Must be called either with bd_mutex held, before a disk can be opened or
320 * after all disk users are gone.
322 static struct block_device *add_partition(struct gendisk *disk, int partno,
323 sector_t start, sector_t len, int flags,
324 struct partition_meta_info *info)
326 dev_t devt = MKDEV(0, 0);
327 struct device *ddev = disk_to_dev(disk);
329 struct block_device *bdev;
330 struct disk_part_tbl *ptbl;
335 * Partitions are not supported on zoned block devices that are used as
338 switch (disk->queue->limits.zoned) {
340 pr_warn("%s: partitions not supported on host managed zoned block device\n",
342 return ERR_PTR(-ENXIO);
344 pr_info("%s: disabling host aware zoned block device support due to partitions\n",
346 disk->queue->limits.zoned = BLK_ZONED_NONE;
352 err = disk_expand_part_tbl(disk, partno);
355 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
357 if (ptbl->part[partno])
358 return ERR_PTR(-EBUSY);
360 bdev = bdev_alloc(disk, partno);
362 return ERR_PTR(-ENOMEM);
364 bdev->bd_start_sect = start;
365 bdev_set_nr_sectors(bdev, len);
366 bdev->bd_read_only = get_disk_ro(disk);
370 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
371 if (!bdev->bd_meta_info)
375 pdev = &bdev->bd_device;
376 dname = dev_name(ddev);
377 if (isdigit(dname[strlen(dname) - 1]))
378 dev_set_name(pdev, "%sp%d", dname, partno);
380 dev_set_name(pdev, "%s%d", dname, partno);
382 device_initialize(pdev);
383 pdev->class = &block_class;
384 pdev->type = &part_type;
387 err = blk_alloc_devt(bdev, &devt);
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 bdev_add(bdev, devt);
412 rcu_assign_pointer(ptbl->part[partno], bdev);
414 /* suppress uevent if the disk suppresses it */
415 if (!dev_get_uevent_suppress(ddev))
416 kobject_uevent(&pdev->kobj, KOBJ_ADD);
423 kobject_put(bdev->bd_holder_dir);
430 static bool partition_overlaps(struct gendisk *disk, sector_t start,
431 sector_t length, int skip_partno)
433 struct disk_part_iter piter;
434 struct block_device *part;
435 bool overlap = false;
437 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
438 while ((part = disk_part_iter_next(&piter))) {
439 if (part->bd_partno == skip_partno ||
440 start >= part->bd_start_sect + bdev_nr_sectors(part) ||
441 start + length <= part->bd_start_sect)
447 disk_part_iter_exit(&piter);
451 int bdev_add_partition(struct block_device *bdev, int partno,
452 sector_t start, sector_t length)
454 struct block_device *part;
456 mutex_lock(&bdev->bd_mutex);
457 if (partition_overlaps(bdev->bd_disk, start, length, -1)) {
458 mutex_unlock(&bdev->bd_mutex);
462 part = add_partition(bdev->bd_disk, partno, start, length,
463 ADDPART_FLAG_NONE, NULL);
464 mutex_unlock(&bdev->bd_mutex);
465 return PTR_ERR_OR_ZERO(part);
468 int bdev_del_partition(struct block_device *bdev, int partno)
470 struct block_device *part;
473 part = bdget_disk(bdev->bd_disk, partno);
477 mutex_lock(&part->bd_mutex);
478 mutex_lock_nested(&bdev->bd_mutex, 1);
481 if (part->bd_openers)
485 invalidate_bdev(part);
487 delete_partition(part);
490 mutex_unlock(&bdev->bd_mutex);
491 mutex_unlock(&part->bd_mutex);
496 int bdev_resize_partition(struct block_device *bdev, int partno,
497 sector_t start, sector_t length)
499 struct block_device *part;
502 part = bdget_disk(bdev->bd_disk, partno);
506 mutex_lock(&part->bd_mutex);
507 mutex_lock_nested(&bdev->bd_mutex, 1);
509 if (start != part->bd_start_sect)
513 if (partition_overlaps(bdev->bd_disk, start, length, partno))
516 bdev_set_nr_sectors(part, length);
520 mutex_unlock(&part->bd_mutex);
521 mutex_unlock(&bdev->bd_mutex);
526 static bool disk_unlock_native_capacity(struct gendisk *disk)
528 const struct block_device_operations *bdops = disk->fops;
530 if (bdops->unlock_native_capacity &&
531 !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
532 printk(KERN_CONT "enabling native capacity\n");
533 bdops->unlock_native_capacity(disk);
534 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
537 printk(KERN_CONT "truncated\n");
542 int blk_drop_partitions(struct block_device *bdev)
544 struct disk_part_iter piter;
545 struct block_device *part;
547 if (bdev->bd_part_count)
551 invalidate_bdev(bdev);
553 disk_part_iter_init(&piter, bdev->bd_disk, DISK_PITER_INCL_EMPTY);
554 while ((part = disk_part_iter_next(&piter)))
555 delete_partition(part);
556 disk_part_iter_exit(&piter);
561 /* for historic reasons in the DASD driver */
562 EXPORT_SYMBOL_GPL(blk_drop_partitions);
565 static bool blk_add_partition(struct gendisk *disk, struct block_device *bdev,
566 struct parsed_partitions *state, int p)
568 sector_t size = state->parts[p].size;
569 sector_t from = state->parts[p].from;
570 struct block_device *part;
575 if (from >= get_capacity(disk)) {
577 "%s: p%d start %llu is beyond EOD, ",
578 disk->disk_name, p, (unsigned long long) from);
579 if (disk_unlock_native_capacity(disk))
584 if (from + size > get_capacity(disk)) {
586 "%s: p%d size %llu extends beyond EOD, ",
587 disk->disk_name, p, (unsigned long long) size);
589 if (disk_unlock_native_capacity(disk))
593 * We can not ignore partitions of broken tables created by for
594 * example camera firmware, but we limit them to the end of the
595 * disk to avoid creating invalid block devices.
597 size = get_capacity(disk) - from;
600 part = add_partition(disk, p, from, size, state->parts[p].flags,
601 &state->parts[p].info);
602 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
603 printk(KERN_ERR " %s: p%d could not be added: %ld\n",
604 disk->disk_name, p, -PTR_ERR(part));
608 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
609 (state->parts[p].flags & ADDPART_FLAG_RAID))
610 md_autodetect_dev(part->bd_dev);
615 int blk_add_partitions(struct gendisk *disk, struct block_device *bdev)
617 struct parsed_partitions *state;
618 int ret = -EAGAIN, p, highest;
620 if (!disk_part_scan_enabled(disk))
623 state = check_partition(disk, bdev);
628 * I/O error reading the partition table. If we tried to read
629 * beyond EOD, retry after unlocking the native capacity.
631 if (PTR_ERR(state) == -ENOSPC) {
632 printk(KERN_WARNING "%s: partition table beyond EOD, ",
634 if (disk_unlock_native_capacity(disk))
641 * Partitions are not supported on host managed zoned block devices.
643 if (disk->queue->limits.zoned == BLK_ZONED_HM) {
644 pr_warn("%s: ignoring partition table on host managed zoned block device\n",
651 * If we read beyond EOD, try unlocking native capacity even if the
652 * partition table was successfully read as we could be missing some
655 if (state->access_beyond_eod) {
657 "%s: partition table partially beyond EOD, ",
659 if (disk_unlock_native_capacity(disk))
663 /* tell userspace that the media / partition table may have changed */
664 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
667 * Detect the highest partition number and preallocate disk->part_tbl.
668 * This is an optimization and not strictly necessary.
670 for (p = 1, highest = 0; p < state->limit; p++)
671 if (state->parts[p].size)
673 disk_expand_part_tbl(disk, highest);
675 for (p = 1; p < state->limit; p++)
676 if (!blk_add_partition(disk, bdev, state, p))
681 free_partitions(state);
685 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
687 struct address_space *mapping = state->bdev->bd_inode->i_mapping;
690 if (n >= get_capacity(state->bdev->bd_disk)) {
691 state->access_beyond_eod = true;
695 page = read_mapping_page(mapping,
696 (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL);
703 return (unsigned char *)page_address(page) +
704 ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT);