1 // SPDX-License-Identifier: GPL-2.0
3 * Code extracted from drivers/block/genhd.c
4 * Copyright (C) 1991-1998 Linus Torvalds
5 * Re-organised Feb 1998 Russell King
7 * We now have independent partition support from the
8 * block drivers, which allows all the partition code to
9 * be grouped in one location, and it to be mostly self
13 #include <linux/init.h>
14 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <linux/kmod.h>
18 #include <linux/ctype.h>
19 #include <linux/genhd.h>
20 #include <linux/blktrace_api.h>
22 #include "partitions/check.h"
24 #ifdef CONFIG_BLK_DEV_MD
25 extern void md_autodetect_dev(dev_t dev);
29 * disk_name() is used by partition check code and the genhd driver.
30 * It formats the devicename of the indicated disk into
31 * the supplied buffer (of size at least 32), and returns
32 * a pointer to that same buffer (for convenience).
35 char *disk_name(struct gendisk *hd, int partno, char *buf)
38 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
39 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
40 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
42 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
47 const char *bdevname(struct block_device *bdev, char *buf)
49 return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
52 EXPORT_SYMBOL(bdevname);
54 const char *bio_devname(struct bio *bio, char *buf)
56 return disk_name(bio->bi_disk, bio->bi_partno, buf);
58 EXPORT_SYMBOL(bio_devname);
61 * There's very little reason to use this, you should really
62 * have a struct block_device just about everywhere and use
65 const char *__bdevname(dev_t dev, char *buffer)
67 scnprintf(buffer, BDEVNAME_SIZE, "unknown-block(%u,%u)",
68 MAJOR(dev), MINOR(dev));
72 EXPORT_SYMBOL(__bdevname);
74 static ssize_t part_partition_show(struct device *dev,
75 struct device_attribute *attr, char *buf)
77 struct hd_struct *p = dev_to_part(dev);
79 return sprintf(buf, "%d\n", p->partno);
82 static ssize_t part_start_show(struct device *dev,
83 struct device_attribute *attr, char *buf)
85 struct hd_struct *p = dev_to_part(dev);
87 return sprintf(buf, "%llu\n",(unsigned long long)p->start_sect);
90 ssize_t part_size_show(struct device *dev,
91 struct device_attribute *attr, char *buf)
93 struct hd_struct *p = dev_to_part(dev);
94 return sprintf(buf, "%llu\n",(unsigned long long)part_nr_sects_read(p));
97 static ssize_t part_ro_show(struct device *dev,
98 struct device_attribute *attr, char *buf)
100 struct hd_struct *p = dev_to_part(dev);
101 return sprintf(buf, "%d\n", p->policy ? 1 : 0);
104 static ssize_t part_alignment_offset_show(struct device *dev,
105 struct device_attribute *attr, char *buf)
107 struct hd_struct *p = dev_to_part(dev);
108 return sprintf(buf, "%llu\n", (unsigned long long)p->alignment_offset);
111 static ssize_t part_discard_alignment_show(struct device *dev,
112 struct device_attribute *attr, char *buf)
114 struct hd_struct *p = dev_to_part(dev);
115 return sprintf(buf, "%u\n", p->discard_alignment);
118 ssize_t part_stat_show(struct device *dev,
119 struct device_attribute *attr, char *buf)
121 struct hd_struct *p = dev_to_part(dev);
122 struct request_queue *q = part_to_disk(p)->queue;
123 unsigned int inflight[2];
126 cpu = part_stat_lock();
127 part_round_stats(q, cpu, p);
129 part_in_flight(q, p, inflight);
131 "%8lu %8lu %8llu %8u "
132 "%8lu %8lu %8llu %8u "
134 "%8lu %8lu %8llu %8u"
136 part_stat_read(p, ios[STAT_READ]),
137 part_stat_read(p, merges[STAT_READ]),
138 (unsigned long long)part_stat_read(p, sectors[STAT_READ]),
139 (unsigned int)part_stat_read_msecs(p, STAT_READ),
140 part_stat_read(p, ios[STAT_WRITE]),
141 part_stat_read(p, merges[STAT_WRITE]),
142 (unsigned long long)part_stat_read(p, sectors[STAT_WRITE]),
143 (unsigned int)part_stat_read_msecs(p, STAT_WRITE),
145 jiffies_to_msecs(part_stat_read(p, io_ticks)),
146 jiffies_to_msecs(part_stat_read(p, time_in_queue)),
147 part_stat_read(p, ios[STAT_DISCARD]),
148 part_stat_read(p, merges[STAT_DISCARD]),
149 (unsigned long long)part_stat_read(p, sectors[STAT_DISCARD]),
150 (unsigned int)part_stat_read_msecs(p, STAT_DISCARD));
153 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
156 struct hd_struct *p = dev_to_part(dev);
157 struct request_queue *q = part_to_disk(p)->queue;
158 unsigned int inflight[2];
160 part_in_flight_rw(q, p, inflight);
161 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
164 #ifdef CONFIG_FAIL_MAKE_REQUEST
165 ssize_t part_fail_show(struct device *dev,
166 struct device_attribute *attr, char *buf)
168 struct hd_struct *p = dev_to_part(dev);
170 return sprintf(buf, "%d\n", p->make_it_fail);
173 ssize_t part_fail_store(struct device *dev,
174 struct device_attribute *attr,
175 const char *buf, size_t count)
177 struct hd_struct *p = dev_to_part(dev);
180 if (count > 0 && sscanf(buf, "%d", &i) > 0)
181 p->make_it_fail = (i == 0) ? 0 : 1;
187 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
188 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
189 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
190 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
191 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
192 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
193 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
194 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
195 #ifdef CONFIG_FAIL_MAKE_REQUEST
196 static struct device_attribute dev_attr_fail =
197 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
200 static struct attribute *part_attrs[] = {
201 &dev_attr_partition.attr,
202 &dev_attr_start.attr,
205 &dev_attr_alignment_offset.attr,
206 &dev_attr_discard_alignment.attr,
208 &dev_attr_inflight.attr,
209 #ifdef CONFIG_FAIL_MAKE_REQUEST
215 static struct attribute_group part_attr_group = {
219 static const struct attribute_group *part_attr_groups[] = {
221 #ifdef CONFIG_BLK_DEV_IO_TRACE
222 &blk_trace_attr_group,
227 static void part_release(struct device *dev)
229 struct hd_struct *p = dev_to_part(dev);
230 blk_free_devt(dev->devt);
235 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
237 struct hd_struct *part = dev_to_part(dev);
239 add_uevent_var(env, "PARTN=%u", part->partno);
240 if (part->info && part->info->volname[0])
241 add_uevent_var(env, "PARTNAME=%s", part->info->volname);
245 struct device_type part_type = {
247 .groups = part_attr_groups,
248 .release = part_release,
249 .uevent = part_uevent,
252 static void delete_partition_work_fn(struct work_struct *work)
254 struct hd_struct *part = container_of(to_rcu_work(work), struct hd_struct,
257 part->start_sect = 0;
259 part_stat_set_all(part, 0);
260 put_device(part_to_dev(part));
263 void __delete_partition(struct percpu_ref *ref)
265 struct hd_struct *part = container_of(ref, struct hd_struct, ref);
266 INIT_RCU_WORK(&part->rcu_work, delete_partition_work_fn);
267 queue_rcu_work(system_wq, &part->rcu_work);
271 * Must be called either with bd_mutex held, before a disk can be opened or
272 * after all disk users are gone.
274 void delete_partition(struct gendisk *disk, int partno)
276 struct disk_part_tbl *ptbl =
277 rcu_dereference_protected(disk->part_tbl, 1);
278 struct hd_struct *part;
279 struct block_device *bdev;
281 if (partno >= ptbl->len)
284 part = rcu_dereference_protected(ptbl->part[partno], 1);
288 rcu_assign_pointer(ptbl->part[partno], NULL);
289 rcu_assign_pointer(ptbl->last_lookup, NULL);
290 kobject_put(part->holder_dir);
291 device_del(part_to_dev(part));
294 * Remove gendisk pointer from idr so that it cannot be looked up
295 * while RCU period before freeing gendisk is running to prevent
296 * use-after-free issues. Note that the device number stays
297 * "in-use" until we really free the gendisk.
299 blk_invalidate_devt(part_devt(part));
301 bdev = bdget(part_devt(part));
303 remove_inode_hash(bdev->bd_inode);
306 hd_struct_kill(part);
309 static ssize_t whole_disk_show(struct device *dev,
310 struct device_attribute *attr, char *buf)
314 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
317 * Must be called either with bd_mutex held, before a disk can be opened or
318 * after all disk users are gone.
320 struct hd_struct *add_partition(struct gendisk *disk, int partno,
321 sector_t start, sector_t len, int flags,
322 struct partition_meta_info *info)
325 dev_t devt = MKDEV(0, 0);
326 struct device *ddev = disk_to_dev(disk);
328 struct disk_part_tbl *ptbl;
332 err = disk_expand_part_tbl(disk, partno);
335 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
337 if (ptbl->part[partno])
338 return ERR_PTR(-EBUSY);
340 p = kzalloc(sizeof(*p), GFP_KERNEL);
342 return ERR_PTR(-EBUSY);
344 if (!init_part_stats(p)) {
349 seqcount_init(&p->nr_sects_seq);
350 pdev = part_to_dev(p);
352 p->start_sect = start;
353 p->alignment_offset =
354 queue_limit_alignment_offset(&disk->queue->limits, start);
355 p->discard_alignment =
356 queue_limit_discard_alignment(&disk->queue->limits, start);
359 p->policy = get_disk_ro(disk);
362 struct partition_meta_info *pinfo = alloc_part_info(disk);
367 memcpy(pinfo, info, sizeof(*info));
371 dname = dev_name(ddev);
372 if (isdigit(dname[strlen(dname) - 1]))
373 dev_set_name(pdev, "%sp%d", dname, partno);
375 dev_set_name(pdev, "%s%d", dname, partno);
377 device_initialize(pdev);
378 pdev->class = &block_class;
379 pdev->type = &part_type;
382 err = blk_alloc_devt(p, &devt);
387 /* delay uevent until 'holders' subdir is created */
388 dev_set_uevent_suppress(pdev, 1);
389 err = device_add(pdev);
394 p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
398 dev_set_uevent_suppress(pdev, 0);
399 if (flags & ADDPART_FLAG_WHOLEDISK) {
400 err = device_create_file(pdev, &dev_attr_whole_disk);
405 err = hd_ref_init(p);
407 if (flags & ADDPART_FLAG_WHOLEDISK)
408 goto out_remove_file;
412 /* everything is up and running, commence */
413 rcu_assign_pointer(ptbl->part[partno], p);
415 /* suppress uevent if the disk suppresses it */
416 if (!dev_get_uevent_suppress(ddev))
417 kobject_uevent(&pdev->kobj, KOBJ_ADD);
428 device_remove_file(pdev, &dev_attr_whole_disk);
430 kobject_put(p->holder_dir);
437 static bool disk_unlock_native_capacity(struct gendisk *disk)
439 const struct block_device_operations *bdops = disk->fops;
441 if (bdops->unlock_native_capacity &&
442 !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
443 printk(KERN_CONT "enabling native capacity\n");
444 bdops->unlock_native_capacity(disk);
445 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
448 printk(KERN_CONT "truncated\n");
453 static int drop_partitions(struct gendisk *disk, struct block_device *bdev)
455 struct disk_part_iter piter;
456 struct hd_struct *part;
459 if (bdev->bd_part_count || bdev->bd_super)
461 res = invalidate_partition(disk, 0);
465 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
466 while ((part = disk_part_iter_next(&piter)))
467 delete_partition(disk, part->partno);
468 disk_part_iter_exit(&piter);
473 static bool part_zone_aligned(struct gendisk *disk,
474 struct block_device *bdev,
475 sector_t from, sector_t size)
477 unsigned int zone_sectors = bdev_zone_sectors(bdev);
480 * If this function is called, then the disk is a zoned block device
481 * (host-aware or host-managed). This can be detected even if the
482 * zoned block device support is disabled (CONFIG_BLK_DEV_ZONED not
483 * set). In this case, however, only host-aware devices will be seen
484 * as a block device is not created for host-managed devices. Without
485 * zoned block device support, host-aware drives can still be used as
486 * regular block devices (no zone operation) and their zone size will
487 * be reported as 0. Allow this case.
493 * Check partition start and size alignement. If the drive has a
494 * smaller last runt zone, ignore it and allow the partition to
495 * use it. Check the zone size too: it should be a power of 2 number
498 if (WARN_ON_ONCE(!is_power_of_2(zone_sectors))) {
501 div_u64_rem(from, zone_sectors, &rem);
504 if ((from + size) < get_capacity(disk)) {
505 div_u64_rem(size, zone_sectors, &rem);
512 if (from & (zone_sectors - 1))
514 if ((from + size) < get_capacity(disk) &&
515 (size & (zone_sectors - 1)))
523 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
525 struct parsed_partitions *state = NULL;
526 struct hd_struct *part;
529 if (state && !IS_ERR(state)) {
530 free_partitions(state);
534 res = drop_partitions(disk, bdev);
538 if (disk->fops->revalidate_disk)
539 disk->fops->revalidate_disk(disk);
540 check_disk_size_change(disk, bdev, true);
541 bdev->bd_invalidated = 0;
542 if (!get_capacity(disk) || !(state = check_partition(disk, bdev)))
546 * I/O error reading the partition table. If any
547 * partition code tried to read beyond EOD, retry
548 * after unlocking native capacity.
550 if (PTR_ERR(state) == -ENOSPC) {
551 printk(KERN_WARNING "%s: partition table beyond EOD, ",
553 if (disk_unlock_native_capacity(disk))
559 * If any partition code tried to read beyond EOD, try
560 * unlocking native capacity even if partition table is
561 * successfully read as we could be missing some partitions.
563 if (state->access_beyond_eod) {
565 "%s: partition table partially beyond EOD, ",
567 if (disk_unlock_native_capacity(disk))
571 /* tell userspace that the media / partition table may have changed */
572 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
574 /* Detect the highest partition number and preallocate
575 * disk->part_tbl. This is an optimization and not strictly
578 for (p = 1, highest = 0; p < state->limit; p++)
579 if (state->parts[p].size)
582 disk_expand_part_tbl(disk, highest);
585 for (p = 1; p < state->limit; p++) {
588 size = state->parts[p].size;
592 from = state->parts[p].from;
593 if (from >= get_capacity(disk)) {
595 "%s: p%d start %llu is beyond EOD, ",
596 disk->disk_name, p, (unsigned long long) from);
597 if (disk_unlock_native_capacity(disk))
602 if (from + size > get_capacity(disk)) {
604 "%s: p%d size %llu extends beyond EOD, ",
605 disk->disk_name, p, (unsigned long long) size);
607 if (disk_unlock_native_capacity(disk)) {
608 /* free state and restart */
612 * we can not ignore partitions of broken tables
613 * created by for example camera firmware, but
614 * we limit them to the end of the disk to avoid
615 * creating invalid block devices
617 size = get_capacity(disk) - from;
622 * On a zoned block device, partitions should be aligned on the
623 * device zone size (i.e. zone boundary crossing not allowed).
624 * Otherwise, resetting the write pointer of the last zone of
625 * one partition may impact the following partition.
627 if (bdev_is_zoned(bdev) &&
628 !part_zone_aligned(disk, bdev, from, size)) {
630 "%s: p%d start %llu+%llu is not zone aligned\n",
631 disk->disk_name, p, (unsigned long long) from,
632 (unsigned long long) size);
636 part = add_partition(disk, p, from, size,
637 state->parts[p].flags,
638 &state->parts[p].info);
640 printk(KERN_ERR " %s: p%d could not be added: %ld\n",
641 disk->disk_name, p, -PTR_ERR(part));
644 #ifdef CONFIG_BLK_DEV_MD
645 if (state->parts[p].flags & ADDPART_FLAG_RAID)
646 md_autodetect_dev(part_to_dev(part)->devt);
649 free_partitions(state);
653 int invalidate_partitions(struct gendisk *disk, struct block_device *bdev)
657 if (!bdev->bd_invalidated)
660 res = drop_partitions(disk, bdev);
664 set_capacity(disk, 0);
665 check_disk_size_change(disk, bdev, false);
666 bdev->bd_invalidated = 0;
667 /* tell userspace that the media / partition table may have changed */
668 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
673 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
675 struct address_space *mapping = bdev->bd_inode->i_mapping;
678 page = read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_SHIFT-9)), NULL);
683 return (unsigned char *)page_address(page) + ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << 9);
691 EXPORT_SYMBOL(read_dev_sector);