1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 1991-1998  Linus Torvalds
4 * Re-organised Feb 1998 Russell King
5 * Copyright (C) 2020 Christoph Hellwig
6 */
7#include <linux/fs.h>
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>
13#include "check.h"
14
15static int (*const check_part[])(struct parsed_partitions *) = {
16	/*
17	 * Probe partition formats with tables at disk address 0
18	 * that also have an ADFS boot block at 0xdc0.
19	 */
20#ifdef CONFIG_ACORN_PARTITION_ICS
21	adfspart_check_ICS,
22#endif
23#ifdef CONFIG_ACORN_PARTITION_POWERTEC
24	adfspart_check_POWERTEC,
25#endif
26#ifdef CONFIG_ACORN_PARTITION_EESOX
27	adfspart_check_EESOX,
28#endif
29
30	/*
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
34	 * the msdos entry.
35	 */
36#ifdef CONFIG_ACORN_PARTITION_CUMANA
37	adfspart_check_CUMANA,
38#endif
39#ifdef CONFIG_ACORN_PARTITION_ADFS
40	adfspart_check_ADFS,
41#endif
42
43#ifdef CONFIG_CMDLINE_PARTITION
44	cmdline_partition,
45#endif
46#ifdef CONFIG_EFI_PARTITION
47	efi_partition,		/* this must come before msdos */
48#endif
49#ifdef CONFIG_SGI_PARTITION
50	sgi_partition,
51#endif
52#ifdef CONFIG_LDM_PARTITION
53	ldm_partition,		/* this must come before msdos */
54#endif
55#ifdef CONFIG_MSDOS_PARTITION
56	msdos_partition,
57#endif
58#ifdef CONFIG_OSF_PARTITION
59	osf_partition,
60#endif
61#ifdef CONFIG_SUN_PARTITION
62	sun_partition,
63#endif
64#ifdef CONFIG_AMIGA_PARTITION
65	amiga_partition,
66#endif
67#ifdef CONFIG_ATARI_PARTITION
68	atari_partition,
69#endif
70#ifdef CONFIG_MAC_PARTITION
71	mac_partition,
72#endif
73#ifdef CONFIG_ULTRIX_PARTITION
74	ultrix_partition,
75#endif
76#ifdef CONFIG_IBM_PARTITION
77	ibm_partition,
78#endif
79#ifdef CONFIG_KARMA_PARTITION
80	karma_partition,
81#endif
82#ifdef CONFIG_SYSV68_PARTITION
83	sysv68_partition,
84#endif
85	NULL
86};
87
88static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
89{
90	struct parsed_partitions *state;
91	int nr = DISK_MAX_PARTS;
92
93	state = kzalloc(sizeof(*state), GFP_KERNEL);
94	if (!state)
95		return NULL;
96
97	state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
98	if (!state->parts) {
99		kfree(state);
100		return NULL;
101	}
102
103	state->limit = nr;
104
105	return state;
106}
107
108static void free_partitions(struct parsed_partitions *state)
109{
110	vfree(state->parts);
111	kfree(state);
112}
113
114static struct parsed_partitions *check_partition(struct gendisk *hd)
115{
116	struct parsed_partitions *state;
117	int i, res, err;
118
119	state = allocate_partitions(hd);
120	if (!state)
121		return NULL;
122	state->pp_buf = (char *)__get_free_page(GFP_KERNEL);
123	if (!state->pp_buf) {
124		free_partitions(state);
125		return NULL;
126	}
127	state->pp_buf[0] = '\0';
128
129	state->disk = hd;
130	snprintf(state->name, BDEVNAME_SIZE, "%s", hd->disk_name);
131	snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name);
132	if (isdigit(state->name[strlen(state->name)-1]))
133		sprintf(state->name, "p");
134
135	i = res = err = 0;
136	while (!res && check_part[i]) {
137		memset(state->parts, 0, state->limit * sizeof(state->parts[0]));
138		res = check_part[i++](state);
139		if (res < 0) {
140			/*
141			 * We have hit an I/O error which we don't report now.
142			 * But record it, and let the others do their job.
143			 */
144			err = res;
145			res = 0;
146		}
147
148	}
149	if (res > 0) {
150		printk(KERN_INFO "%s", state->pp_buf);
151
152		free_page((unsigned long)state->pp_buf);
153		return state;
154	}
155	if (state->access_beyond_eod)
156		err = -ENOSPC;
157	/*
158	 * The partition is unrecognized. So report I/O errors if there were any
159	 */
160	if (err)
161		res = err;
162	if (res) {
163		strlcat(state->pp_buf,
164			" unable to read partition table\n", PAGE_SIZE);
165		printk(KERN_INFO "%s", state->pp_buf);
166	}
167
168	free_page((unsigned long)state->pp_buf);
169	free_partitions(state);
170	return ERR_PTR(res);
171}
172
173static ssize_t part_partition_show(struct device *dev,
174				   struct device_attribute *attr, char *buf)
175{
176	return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno);
177}
178
179static ssize_t part_start_show(struct device *dev,
180			       struct device_attribute *attr, char *buf)
181{
182	return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect);
183}
184
185static ssize_t part_ro_show(struct device *dev,
186			    struct device_attribute *attr, char *buf)
187{
188	return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev)));
189}
190
191static ssize_t part_alignment_offset_show(struct device *dev,
192					  struct device_attribute *attr, char *buf)
193{
194	return sprintf(buf, "%u\n", bdev_alignment_offset(dev_to_bdev(dev)));
195}
196
197static ssize_t part_discard_alignment_show(struct device *dev,
198					   struct device_attribute *attr, char *buf)
199{
200	return sprintf(buf, "%u\n", bdev_discard_alignment(dev_to_bdev(dev)));
201}
202
203static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
204static DEVICE_ATTR(start, 0444, part_start_show, NULL);
205static DEVICE_ATTR(size, 0444, part_size_show, NULL);
206static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
207static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
208static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
209static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
210static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
211#ifdef CONFIG_FAIL_MAKE_REQUEST
212static struct device_attribute dev_attr_fail =
213	__ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
214#endif
215
216static struct attribute *part_attrs[] = {
217	&dev_attr_partition.attr,
218	&dev_attr_start.attr,
219	&dev_attr_size.attr,
220	&dev_attr_ro.attr,
221	&dev_attr_alignment_offset.attr,
222	&dev_attr_discard_alignment.attr,
223	&dev_attr_stat.attr,
224	&dev_attr_inflight.attr,
225#ifdef CONFIG_FAIL_MAKE_REQUEST
226	&dev_attr_fail.attr,
227#endif
228	NULL
229};
230
231static const struct attribute_group part_attr_group = {
232	.attrs = part_attrs,
233};
234
235static const struct attribute_group *part_attr_groups[] = {
236	&part_attr_group,
237#ifdef CONFIG_BLK_DEV_IO_TRACE
238	&blk_trace_attr_group,
239#endif
240	NULL
241};
242
243static void part_release(struct device *dev)
244{
245	put_disk(dev_to_bdev(dev)->bd_disk);
246	iput(dev_to_bdev(dev)->bd_inode);
247}
248
249static int part_uevent(const struct device *dev, struct kobj_uevent_env *env)
250{
251	const struct block_device *part = dev_to_bdev(dev);
252
253	add_uevent_var(env, "PARTN=%u", part->bd_partno);
254	if (part->bd_meta_info && part->bd_meta_info->volname[0])
255		add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname);
256	return 0;
257}
258
259const struct device_type part_type = {
260	.name		= "partition",
261	.groups		= part_attr_groups,
262	.release	= part_release,
263	.uevent		= part_uevent,
264};
265
266void drop_partition(struct block_device *part)
267{
268	lockdep_assert_held(&part->bd_disk->open_mutex);
269
270	xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
271	kobject_put(part->bd_holder_dir);
272
273	device_del(&part->bd_device);
274	put_device(&part->bd_device);
275}
276
277static ssize_t whole_disk_show(struct device *dev,
278			       struct device_attribute *attr, char *buf)
279{
280	return 0;
281}
282static const DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
283
284/*
285 * Must be called either with open_mutex held, before a disk can be opened or
286 * after all disk users are gone.
287 */
288static struct block_device *add_partition(struct gendisk *disk, int partno,
289				sector_t start, sector_t len, int flags,
290				struct partition_meta_info *info)
291{
292	dev_t devt = MKDEV(0, 0);
293	struct device *ddev = disk_to_dev(disk);
294	struct device *pdev;
295	struct block_device *bdev;
296	const char *dname;
297	int err;
298
299	lockdep_assert_held(&disk->open_mutex);
300
301	if (partno >= DISK_MAX_PARTS)
302		return ERR_PTR(-EINVAL);
303
304	/*
305	 * Partitions are not supported on zoned block devices that are used as
306	 * such.
307	 */
308	if (bdev_is_zoned(disk->part0)) {
309		pr_warn("%s: partitions not supported on host managed zoned block device\n",
310			disk->disk_name);
311		return ERR_PTR(-ENXIO);
312	}
313
314	if (xa_load(&disk->part_tbl, partno))
315		return ERR_PTR(-EBUSY);
316
317	/* ensure we always have a reference to the whole disk */
318	get_device(disk_to_dev(disk));
319
320	err = -ENOMEM;
321	bdev = bdev_alloc(disk, partno);
322	if (!bdev)
323		goto out_put_disk;
324
325	bdev->bd_start_sect = start;
326	bdev_set_nr_sectors(bdev, len);
327
328	pdev = &bdev->bd_device;
329	dname = dev_name(ddev);
330	if (isdigit(dname[strlen(dname) - 1]))
331		dev_set_name(pdev, "%sp%d", dname, partno);
332	else
333		dev_set_name(pdev, "%s%d", dname, partno);
334
335	device_initialize(pdev);
336	pdev->class = &block_class;
337	pdev->type = &part_type;
338	pdev->parent = ddev;
339
340	/* in consecutive minor range? */
341	if (bdev->bd_partno < disk->minors) {
342		devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno);
343	} else {
344		err = blk_alloc_ext_minor();
345		if (err < 0)
346			goto out_put;
347		devt = MKDEV(BLOCK_EXT_MAJOR, err);
348	}
349	pdev->devt = devt;
350
351	if (info) {
352		err = -ENOMEM;
353		bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
354		if (!bdev->bd_meta_info)
355			goto out_put;
356	}
357
358	/* delay uevent until 'holders' subdir is created */
359	dev_set_uevent_suppress(pdev, 1);
360	err = device_add(pdev);
361	if (err)
362		goto out_put;
363
364	err = -ENOMEM;
365	bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
366	if (!bdev->bd_holder_dir)
367		goto out_del;
368
369	dev_set_uevent_suppress(pdev, 0);
370	if (flags & ADDPART_FLAG_WHOLEDISK) {
371		err = device_create_file(pdev, &dev_attr_whole_disk);
372		if (err)
373			goto out_del;
374	}
375
376	/* everything is up and running, commence */
377	err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
378	if (err)
379		goto out_del;
380	bdev_add(bdev, devt);
381
382	/* suppress uevent if the disk suppresses it */
383	if (!dev_get_uevent_suppress(ddev))
384		kobject_uevent(&pdev->kobj, KOBJ_ADD);
385	return bdev;
386
387out_del:
388	kobject_put(bdev->bd_holder_dir);
389	device_del(pdev);
390out_put:
391	put_device(pdev);
392	return ERR_PTR(err);
393out_put_disk:
394	put_disk(disk);
395	return ERR_PTR(err);
396}
397
398static bool partition_overlaps(struct gendisk *disk, sector_t start,
399		sector_t length, int skip_partno)
400{
401	struct block_device *part;
402	bool overlap = false;
403	unsigned long idx;
404
405	rcu_read_lock();
406	xa_for_each_start(&disk->part_tbl, idx, part, 1) {
407		if (part->bd_partno != skip_partno &&
408		    start < part->bd_start_sect + bdev_nr_sectors(part) &&
409		    start + length > part->bd_start_sect) {
410			overlap = true;
411			break;
412		}
413	}
414	rcu_read_unlock();
415
416	return overlap;
417}
418
419int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
420		sector_t length)
421{
422	struct block_device *part;
423	int ret;
424
425	mutex_lock(&disk->open_mutex);
426	if (!disk_live(disk)) {
427		ret = -ENXIO;
428		goto out;
429	}
430
431	if (disk->flags & GENHD_FL_NO_PART) {
432		ret = -EINVAL;
433		goto out;
434	}
435
436	if (partition_overlaps(disk, start, length, -1)) {
437		ret = -EBUSY;
438		goto out;
439	}
440
441	part = add_partition(disk, partno, start, length,
442			ADDPART_FLAG_NONE, NULL);
443	ret = PTR_ERR_OR_ZERO(part);
444out:
445	mutex_unlock(&disk->open_mutex);
446	return ret;
447}
448
449int bdev_del_partition(struct gendisk *disk, int partno)
450{
451	struct block_device *part = NULL;
452	int ret = -ENXIO;
453
454	mutex_lock(&disk->open_mutex);
455	part = xa_load(&disk->part_tbl, partno);
456	if (!part)
457		goto out_unlock;
458
459	ret = -EBUSY;
460	if (atomic_read(&part->bd_openers))
461		goto out_unlock;
462
463	/*
464	 * We verified that @part->bd_openers is zero above and so
465	 * @part->bd_holder{_ops} can't be set. And since we hold
466	 * @disk->open_mutex the device can't be claimed by anyone.
467	 *
468	 * So no need to call @part->bd_holder_ops->mark_dead() here.
469	 * Just delete the partition and invalidate it.
470	 */
471
472	remove_inode_hash(part->bd_inode);
473	invalidate_bdev(part);
474	drop_partition(part);
475	ret = 0;
476out_unlock:
477	mutex_unlock(&disk->open_mutex);
478	return ret;
479}
480
481int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
482		sector_t length)
483{
484	struct block_device *part = NULL;
485	int ret = -ENXIO;
486
487	mutex_lock(&disk->open_mutex);
488	part = xa_load(&disk->part_tbl, partno);
489	if (!part)
490		goto out_unlock;
491
492	ret = -EINVAL;
493	if (start != part->bd_start_sect)
494		goto out_unlock;
495
496	ret = -EBUSY;
497	if (partition_overlaps(disk, start, length, partno))
498		goto out_unlock;
499
500	bdev_set_nr_sectors(part, length);
501
502	ret = 0;
503out_unlock:
504	mutex_unlock(&disk->open_mutex);
505	return ret;
506}
507
508static bool disk_unlock_native_capacity(struct gendisk *disk)
509{
510	if (!disk->fops->unlock_native_capacity ||
511	    test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) {
512		printk(KERN_CONT "truncated\n");
513		return false;
514	}
515
516	printk(KERN_CONT "enabling native capacity\n");
517	disk->fops->unlock_native_capacity(disk);
518	return true;
519}
520
521static bool blk_add_partition(struct gendisk *disk,
522		struct parsed_partitions *state, int p)
523{
524	sector_t size = state->parts[p].size;
525	sector_t from = state->parts[p].from;
526	struct block_device *part;
527
528	if (!size)
529		return true;
530
531	if (from >= get_capacity(disk)) {
532		printk(KERN_WARNING
533		       "%s: p%d start %llu is beyond EOD, ",
534		       disk->disk_name, p, (unsigned long long) from);
535		if (disk_unlock_native_capacity(disk))
536			return false;
537		return true;
538	}
539
540	if (from + size > get_capacity(disk)) {
541		printk(KERN_WARNING
542		       "%s: p%d size %llu extends beyond EOD, ",
543		       disk->disk_name, p, (unsigned long long) size);
544
545		if (disk_unlock_native_capacity(disk))
546			return false;
547
548		/*
549		 * We can not ignore partitions of broken tables created by for
550		 * example camera firmware, but we limit them to the end of the
551		 * disk to avoid creating invalid block devices.
552		 */
553		size = get_capacity(disk) - from;
554	}
555
556	part = add_partition(disk, p, from, size, state->parts[p].flags,
557			     &state->parts[p].info);
558	if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
559		printk(KERN_ERR " %s: p%d could not be added: %pe\n",
560		       disk->disk_name, p, part);
561		return true;
562	}
563
564	if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
565	    (state->parts[p].flags & ADDPART_FLAG_RAID))
566		md_autodetect_dev(part->bd_dev);
567
568	return true;
569}
570
571static int blk_add_partitions(struct gendisk *disk)
572{
573	struct parsed_partitions *state;
574	int ret = -EAGAIN, p;
575
576	if (disk->flags & GENHD_FL_NO_PART)
577		return 0;
578
579	if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
580		return 0;
581
582	state = check_partition(disk);
583	if (!state)
584		return 0;
585	if (IS_ERR(state)) {
586		/*
587		 * I/O error reading the partition table.  If we tried to read
588		 * beyond EOD, retry after unlocking the native capacity.
589		 */
590		if (PTR_ERR(state) == -ENOSPC) {
591			printk(KERN_WARNING "%s: partition table beyond EOD, ",
592			       disk->disk_name);
593			if (disk_unlock_native_capacity(disk))
594				return -EAGAIN;
595		}
596		return -EIO;
597	}
598
599	/*
600	 * Partitions are not supported on host managed zoned block devices.
601	 */
602	if (bdev_is_zoned(disk->part0)) {
603		pr_warn("%s: ignoring partition table on host managed zoned block device\n",
604			disk->disk_name);
605		ret = 0;
606		goto out_free_state;
607	}
608
609	/*
610	 * If we read beyond EOD, try unlocking native capacity even if the
611	 * partition table was successfully read as we could be missing some
612	 * partitions.
613	 */
614	if (state->access_beyond_eod) {
615		printk(KERN_WARNING
616		       "%s: partition table partially beyond EOD, ",
617		       disk->disk_name);
618		if (disk_unlock_native_capacity(disk))
619			goto out_free_state;
620	}
621
622	/* tell userspace that the media / partition table may have changed */
623	kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
624
625	for (p = 1; p < state->limit; p++)
626		if (!blk_add_partition(disk, state, p))
627			goto out_free_state;
628
629	ret = 0;
630out_free_state:
631	free_partitions(state);
632	return ret;
633}
634
635int bdev_disk_changed(struct gendisk *disk, bool invalidate)
636{
637	struct block_device *part;
638	unsigned long idx;
639	int ret = 0;
640
641	lockdep_assert_held(&disk->open_mutex);
642
643	if (!disk_live(disk))
644		return -ENXIO;
645
646rescan:
647	if (disk->open_partitions)
648		return -EBUSY;
649	sync_blockdev(disk->part0);
650	invalidate_bdev(disk->part0);
651
652	xa_for_each_start(&disk->part_tbl, idx, part, 1) {
653		/*
654		 * Remove the block device from the inode hash, so that
655		 * it cannot be looked up any more even when openers
656		 * still hold references.
657		 */
658		remove_inode_hash(part->bd_inode);
659
660		/*
661		 * If @disk->open_partitions isn't elevated but there's
662		 * still an active holder of that block device things
663		 * are broken.
664		 */
665		WARN_ON_ONCE(atomic_read(&part->bd_openers));
666		invalidate_bdev(part);
667		drop_partition(part);
668	}
669	clear_bit(GD_NEED_PART_SCAN, &disk->state);
670
671	/*
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.
678	 */
679	if (invalidate) {
680		if (!(disk->flags & GENHD_FL_NO_PART) ||
681		    !(disk->flags & GENHD_FL_REMOVABLE))
682			set_capacity(disk, 0);
683	}
684
685	if (get_capacity(disk)) {
686		ret = blk_add_partitions(disk);
687		if (ret == -EAGAIN)
688			goto rescan;
689	} else if (invalidate) {
690		/*
691		 * Tell userspace that the media / partition table may have
692		 * changed.
693		 */
694		kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
695	}
696
697	return ret;
698}
699/*
700 * Only exported for loop and dasd for historic reasons.  Don't use in new
701 * code!
702 */
703EXPORT_SYMBOL_GPL(bdev_disk_changed);
704
705void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
706{
707	struct address_space *mapping = state->disk->part0->bd_inode->i_mapping;
708	struct folio *folio;
709
710	if (n >= get_capacity(state->disk)) {
711		state->access_beyond_eod = true;
712		goto out;
713	}
714
715	folio = read_mapping_folio(mapping, n >> PAGE_SECTORS_SHIFT, NULL);
716	if (IS_ERR(folio))
717		goto out;
718
719	p->v = folio;
720	return folio_address(folio) + offset_in_folio(folio, n * SECTOR_SIZE);
721out:
722	p->v = NULL;
723	return NULL;
724}
725