xref: /linux/block/partitions/core.c (revision c79c3c34f75d72a066e292b10aa50fc758c97c89)
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/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>
14 #include "check.h"
15 
16 static int (*check_part[])(struct parsed_partitions *) = {
17 	/*
18 	 * Probe partition formats with tables at disk address 0
19 	 * that also have an ADFS boot block at 0xdc0.
20 	 */
21 #ifdef CONFIG_ACORN_PARTITION_ICS
22 	adfspart_check_ICS,
23 #endif
24 #ifdef CONFIG_ACORN_PARTITION_POWERTEC
25 	adfspart_check_POWERTEC,
26 #endif
27 #ifdef CONFIG_ACORN_PARTITION_EESOX
28 	adfspart_check_EESOX,
29 #endif
30 
31 	/*
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
35 	 * the msdos entry.
36 	 */
37 #ifdef CONFIG_ACORN_PARTITION_CUMANA
38 	adfspart_check_CUMANA,
39 #endif
40 #ifdef CONFIG_ACORN_PARTITION_ADFS
41 	adfspart_check_ADFS,
42 #endif
43 
44 #ifdef CONFIG_CMDLINE_PARTITION
45 	cmdline_partition,
46 #endif
47 #ifdef CONFIG_EFI_PARTITION
48 	efi_partition,		/* this must come before msdos */
49 #endif
50 #ifdef CONFIG_SGI_PARTITION
51 	sgi_partition,
52 #endif
53 #ifdef CONFIG_LDM_PARTITION
54 	ldm_partition,		/* this must come before msdos */
55 #endif
56 #ifdef CONFIG_MSDOS_PARTITION
57 	msdos_partition,
58 #endif
59 #ifdef CONFIG_OSF_PARTITION
60 	osf_partition,
61 #endif
62 #ifdef CONFIG_SUN_PARTITION
63 	sun_partition,
64 #endif
65 #ifdef CONFIG_AMIGA_PARTITION
66 	amiga_partition,
67 #endif
68 #ifdef CONFIG_ATARI_PARTITION
69 	atari_partition,
70 #endif
71 #ifdef CONFIG_MAC_PARTITION
72 	mac_partition,
73 #endif
74 #ifdef CONFIG_ULTRIX_PARTITION
75 	ultrix_partition,
76 #endif
77 #ifdef CONFIG_IBM_PARTITION
78 	ibm_partition,
79 #endif
80 #ifdef CONFIG_KARMA_PARTITION
81 	karma_partition,
82 #endif
83 #ifdef CONFIG_SYSV68_PARTITION
84 	sysv68_partition,
85 #endif
86 	NULL
87 };
88 
89 static void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
90 {
91 	unsigned long flags;
92 
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);
96 }
97 
98 static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
99 {
100 	struct parsed_partitions *state;
101 	int nr;
102 
103 	state = kzalloc(sizeof(*state), GFP_KERNEL);
104 	if (!state)
105 		return NULL;
106 
107 	nr = disk_max_parts(hd);
108 	state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
109 	if (!state->parts) {
110 		kfree(state);
111 		return NULL;
112 	}
113 
114 	state->limit = nr;
115 
116 	return state;
117 }
118 
119 static void free_partitions(struct parsed_partitions *state)
120 {
121 	vfree(state->parts);
122 	kfree(state);
123 }
124 
125 static struct parsed_partitions *check_partition(struct gendisk *hd,
126 		struct block_device *bdev)
127 {
128 	struct parsed_partitions *state;
129 	int i, res, err;
130 
131 	state = allocate_partitions(hd);
132 	if (!state)
133 		return NULL;
134 	state->pp_buf = (char *)__get_free_page(GFP_KERNEL);
135 	if (!state->pp_buf) {
136 		free_partitions(state);
137 		return NULL;
138 	}
139 	state->pp_buf[0] = '\0';
140 
141 	state->bdev = bdev;
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");
146 
147 	i = res = err = 0;
148 	while (!res && check_part[i]) {
149 		memset(state->parts, 0, state->limit * sizeof(state->parts[0]));
150 		res = check_part[i++](state);
151 		if (res < 0) {
152 			/*
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.
155 			 */
156 			err = res;
157 			res = 0;
158 		}
159 
160 	}
161 	if (res > 0) {
162 		printk(KERN_INFO "%s", state->pp_buf);
163 
164 		free_page((unsigned long)state->pp_buf);
165 		return state;
166 	}
167 	if (state->access_beyond_eod)
168 		err = -ENOSPC;
169 	/*
170 	 * The partition is unrecognized. So report I/O errors if there were any
171 	 */
172 	if (err)
173 		res = err;
174 	if (res) {
175 		strlcat(state->pp_buf,
176 			" unable to read partition table\n", PAGE_SIZE);
177 		printk(KERN_INFO "%s", state->pp_buf);
178 	}
179 
180 	free_page((unsigned long)state->pp_buf);
181 	free_partitions(state);
182 	return ERR_PTR(res);
183 }
184 
185 static ssize_t part_partition_show(struct device *dev,
186 				   struct device_attribute *attr, char *buf)
187 {
188 	return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno);
189 }
190 
191 static ssize_t part_start_show(struct device *dev,
192 			       struct device_attribute *attr, char *buf)
193 {
194 	return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect);
195 }
196 
197 static ssize_t part_ro_show(struct device *dev,
198 			    struct device_attribute *attr, char *buf)
199 {
200 	return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev)));
201 }
202 
203 static ssize_t part_alignment_offset_show(struct device *dev,
204 					  struct device_attribute *attr, char *buf)
205 {
206 	struct block_device *bdev = dev_to_bdev(dev);
207 
208 	return sprintf(buf, "%u\n",
209 		queue_limit_alignment_offset(&bdev->bd_disk->queue->limits,
210 				bdev->bd_start_sect));
211 }
212 
213 static ssize_t part_discard_alignment_show(struct device *dev,
214 					   struct device_attribute *attr, char *buf)
215 {
216 	struct block_device *bdev = dev_to_bdev(dev);
217 
218 	return sprintf(buf, "%u\n",
219 		queue_limit_discard_alignment(&bdev->bd_disk->queue->limits,
220 				bdev->bd_start_sect));
221 }
222 
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);
234 #endif
235 
236 static struct attribute *part_attrs[] = {
237 	&dev_attr_partition.attr,
238 	&dev_attr_start.attr,
239 	&dev_attr_size.attr,
240 	&dev_attr_ro.attr,
241 	&dev_attr_alignment_offset.attr,
242 	&dev_attr_discard_alignment.attr,
243 	&dev_attr_stat.attr,
244 	&dev_attr_inflight.attr,
245 #ifdef CONFIG_FAIL_MAKE_REQUEST
246 	&dev_attr_fail.attr,
247 #endif
248 	NULL
249 };
250 
251 static struct attribute_group part_attr_group = {
252 	.attrs = part_attrs,
253 };
254 
255 static const struct attribute_group *part_attr_groups[] = {
256 	&part_attr_group,
257 #ifdef CONFIG_BLK_DEV_IO_TRACE
258 	&blk_trace_attr_group,
259 #endif
260 	NULL
261 };
262 
263 static void part_release(struct device *dev)
264 {
265 	blk_free_devt(dev->devt);
266 	bdput(dev_to_bdev(dev));
267 }
268 
269 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
270 {
271 	struct block_device *part = dev_to_bdev(dev);
272 
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);
276 	return 0;
277 }
278 
279 struct device_type part_type = {
280 	.name		= "partition",
281 	.groups		= part_attr_groups,
282 	.release	= part_release,
283 	.uevent		= part_uevent,
284 };
285 
286 /*
287  * Must be called either with bd_mutex held, before a disk can be opened or
288  * after all disk users are gone.
289  */
290 void delete_partition(struct block_device *part)
291 {
292 	xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
293 	kobject_put(part->bd_holder_dir);
294 	device_del(&part->bd_device);
295 
296 	/*
297 	 * Remove the block device from the inode hash, so that it cannot be
298 	 * looked up any more even when openers still hold references.
299 	 */
300 	remove_inode_hash(part->bd_inode);
301 
302 	put_device(&part->bd_device);
303 }
304 
305 static ssize_t whole_disk_show(struct device *dev,
306 			       struct device_attribute *attr, char *buf)
307 {
308 	return 0;
309 }
310 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
311 
312 /*
313  * Must be called either with bd_mutex held, before a disk can be opened or
314  * after all disk users are gone.
315  */
316 static struct block_device *add_partition(struct gendisk *disk, int partno,
317 				sector_t start, sector_t len, int flags,
318 				struct partition_meta_info *info)
319 {
320 	dev_t devt = MKDEV(0, 0);
321 	struct device *ddev = disk_to_dev(disk);
322 	struct device *pdev;
323 	struct block_device *bdev;
324 	const char *dname;
325 	int err;
326 
327 	/*
328 	 * Partitions are not supported on zoned block devices that are used as
329 	 * such.
330 	 */
331 	switch (disk->queue->limits.zoned) {
332 	case BLK_ZONED_HM:
333 		pr_warn("%s: partitions not supported on host managed zoned block device\n",
334 			disk->disk_name);
335 		return ERR_PTR(-ENXIO);
336 	case BLK_ZONED_HA:
337 		pr_info("%s: disabling host aware zoned block device support due to partitions\n",
338 			disk->disk_name);
339 		blk_queue_set_zoned(disk, BLK_ZONED_NONE);
340 		break;
341 	case BLK_ZONED_NONE:
342 		break;
343 	}
344 
345 	if (xa_load(&disk->part_tbl, partno))
346 		return ERR_PTR(-EBUSY);
347 
348 	bdev = bdev_alloc(disk, partno);
349 	if (!bdev)
350 		return ERR_PTR(-ENOMEM);
351 
352 	bdev->bd_start_sect = start;
353 	bdev_set_nr_sectors(bdev, len);
354 
355 	if (info) {
356 		err = -ENOMEM;
357 		bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
358 		if (!bdev->bd_meta_info)
359 			goto out_bdput;
360 	}
361 
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);
366 	else
367 		dev_set_name(pdev, "%s%d", dname, partno);
368 
369 	device_initialize(pdev);
370 	pdev->class = &block_class;
371 	pdev->type = &part_type;
372 	pdev->parent = ddev;
373 
374 	err = blk_alloc_devt(bdev, &devt);
375 	if (err)
376 		goto out_put;
377 	pdev->devt = devt;
378 
379 	/* delay uevent until 'holders' subdir is created */
380 	dev_set_uevent_suppress(pdev, 1);
381 	err = device_add(pdev);
382 	if (err)
383 		goto out_put;
384 
385 	err = -ENOMEM;
386 	bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
387 	if (!bdev->bd_holder_dir)
388 		goto out_del;
389 
390 	dev_set_uevent_suppress(pdev, 0);
391 	if (flags & ADDPART_FLAG_WHOLEDISK) {
392 		err = device_create_file(pdev, &dev_attr_whole_disk);
393 		if (err)
394 			goto out_del;
395 	}
396 
397 	/* everything is up and running, commence */
398 	err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
399 	if (err)
400 		goto out_del;
401 	bdev_add(bdev, devt);
402 
403 	/* suppress uevent if the disk suppresses it */
404 	if (!dev_get_uevent_suppress(ddev))
405 		kobject_uevent(&pdev->kobj, KOBJ_ADD);
406 	return bdev;
407 
408 out_bdput:
409 	bdput(bdev);
410 	return ERR_PTR(err);
411 out_del:
412 	kobject_put(bdev->bd_holder_dir);
413 	device_del(pdev);
414 out_put:
415 	put_device(pdev);
416 	return ERR_PTR(err);
417 }
418 
419 static bool partition_overlaps(struct gendisk *disk, sector_t start,
420 		sector_t length, int skip_partno)
421 {
422 	struct disk_part_iter piter;
423 	struct block_device *part;
424 	bool overlap = false;
425 
426 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
427 	while ((part = disk_part_iter_next(&piter))) {
428 		if (part->bd_partno == skip_partno ||
429 		    start >= part->bd_start_sect + bdev_nr_sectors(part) ||
430 		    start + length <= part->bd_start_sect)
431 			continue;
432 		overlap = true;
433 		break;
434 	}
435 
436 	disk_part_iter_exit(&piter);
437 	return overlap;
438 }
439 
440 int bdev_add_partition(struct block_device *bdev, int partno,
441 		sector_t start, sector_t length)
442 {
443 	struct block_device *part;
444 
445 	mutex_lock(&bdev->bd_mutex);
446 	if (partition_overlaps(bdev->bd_disk, start, length, -1)) {
447 		mutex_unlock(&bdev->bd_mutex);
448 		return -EBUSY;
449 	}
450 
451 	part = add_partition(bdev->bd_disk, partno, start, length,
452 			ADDPART_FLAG_NONE, NULL);
453 	mutex_unlock(&bdev->bd_mutex);
454 	return PTR_ERR_OR_ZERO(part);
455 }
456 
457 int bdev_del_partition(struct block_device *bdev, int partno)
458 {
459 	struct block_device *part;
460 	int ret;
461 
462 	part = bdget_disk(bdev->bd_disk, partno);
463 	if (!part)
464 		return -ENXIO;
465 
466 	mutex_lock(&part->bd_mutex);
467 	mutex_lock_nested(&bdev->bd_mutex, 1);
468 
469 	ret = -EBUSY;
470 	if (part->bd_openers)
471 		goto out_unlock;
472 
473 	sync_blockdev(part);
474 	invalidate_bdev(part);
475 
476 	delete_partition(part);
477 	ret = 0;
478 out_unlock:
479 	mutex_unlock(&bdev->bd_mutex);
480 	mutex_unlock(&part->bd_mutex);
481 	bdput(part);
482 	return ret;
483 }
484 
485 int bdev_resize_partition(struct block_device *bdev, int partno,
486 		sector_t start, sector_t length)
487 {
488 	struct block_device *part;
489 	int ret = 0;
490 
491 	part = bdget_disk(bdev->bd_disk, partno);
492 	if (!part)
493 		return -ENXIO;
494 
495 	mutex_lock(&part->bd_mutex);
496 	mutex_lock_nested(&bdev->bd_mutex, 1);
497 	ret = -EINVAL;
498 	if (start != part->bd_start_sect)
499 		goto out_unlock;
500 
501 	ret = -EBUSY;
502 	if (partition_overlaps(bdev->bd_disk, start, length, partno))
503 		goto out_unlock;
504 
505 	bdev_set_nr_sectors(part, length);
506 
507 	ret = 0;
508 out_unlock:
509 	mutex_unlock(&part->bd_mutex);
510 	mutex_unlock(&bdev->bd_mutex);
511 	bdput(part);
512 	return ret;
513 }
514 
515 static bool disk_unlock_native_capacity(struct gendisk *disk)
516 {
517 	const struct block_device_operations *bdops = disk->fops;
518 
519 	if (bdops->unlock_native_capacity &&
520 	    !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
521 		printk(KERN_CONT "enabling native capacity\n");
522 		bdops->unlock_native_capacity(disk);
523 		disk->flags |= GENHD_FL_NATIVE_CAPACITY;
524 		return true;
525 	} else {
526 		printk(KERN_CONT "truncated\n");
527 		return false;
528 	}
529 }
530 
531 int blk_drop_partitions(struct block_device *bdev)
532 {
533 	struct disk_part_iter piter;
534 	struct block_device *part;
535 
536 	if (bdev->bd_part_count)
537 		return -EBUSY;
538 
539 	sync_blockdev(bdev);
540 	invalidate_bdev(bdev);
541 
542 	disk_part_iter_init(&piter, bdev->bd_disk, DISK_PITER_INCL_EMPTY);
543 	while ((part = disk_part_iter_next(&piter)))
544 		delete_partition(part);
545 	disk_part_iter_exit(&piter);
546 
547 	return 0;
548 }
549 #ifdef CONFIG_S390
550 /* for historic reasons in the DASD driver */
551 EXPORT_SYMBOL_GPL(blk_drop_partitions);
552 #endif
553 
554 static bool blk_add_partition(struct gendisk *disk, struct block_device *bdev,
555 		struct parsed_partitions *state, int p)
556 {
557 	sector_t size = state->parts[p].size;
558 	sector_t from = state->parts[p].from;
559 	struct block_device *part;
560 
561 	if (!size)
562 		return true;
563 
564 	if (from >= get_capacity(disk)) {
565 		printk(KERN_WARNING
566 		       "%s: p%d start %llu is beyond EOD, ",
567 		       disk->disk_name, p, (unsigned long long) from);
568 		if (disk_unlock_native_capacity(disk))
569 			return false;
570 		return true;
571 	}
572 
573 	if (from + size > get_capacity(disk)) {
574 		printk(KERN_WARNING
575 		       "%s: p%d size %llu extends beyond EOD, ",
576 		       disk->disk_name, p, (unsigned long long) size);
577 
578 		if (disk_unlock_native_capacity(disk))
579 			return false;
580 
581 		/*
582 		 * We can not ignore partitions of broken tables created by for
583 		 * example camera firmware, but we limit them to the end of the
584 		 * disk to avoid creating invalid block devices.
585 		 */
586 		size = get_capacity(disk) - from;
587 	}
588 
589 	part = add_partition(disk, p, from, size, state->parts[p].flags,
590 			     &state->parts[p].info);
591 	if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
592 		printk(KERN_ERR " %s: p%d could not be added: %ld\n",
593 		       disk->disk_name, p, -PTR_ERR(part));
594 		return true;
595 	}
596 
597 	if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
598 	    (state->parts[p].flags & ADDPART_FLAG_RAID))
599 		md_autodetect_dev(part->bd_dev);
600 
601 	return true;
602 }
603 
604 int blk_add_partitions(struct gendisk *disk, struct block_device *bdev)
605 {
606 	struct parsed_partitions *state;
607 	int ret = -EAGAIN, p;
608 
609 	if (!disk_part_scan_enabled(disk))
610 		return 0;
611 
612 	state = check_partition(disk, bdev);
613 	if (!state)
614 		return 0;
615 	if (IS_ERR(state)) {
616 		/*
617 		 * I/O error reading the partition table.  If we tried to read
618 		 * beyond EOD, retry after unlocking the native capacity.
619 		 */
620 		if (PTR_ERR(state) == -ENOSPC) {
621 			printk(KERN_WARNING "%s: partition table beyond EOD, ",
622 			       disk->disk_name);
623 			if (disk_unlock_native_capacity(disk))
624 				return -EAGAIN;
625 		}
626 		return -EIO;
627 	}
628 
629 	/*
630 	 * Partitions are not supported on host managed zoned block devices.
631 	 */
632 	if (disk->queue->limits.zoned == BLK_ZONED_HM) {
633 		pr_warn("%s: ignoring partition table on host managed zoned block device\n",
634 			disk->disk_name);
635 		ret = 0;
636 		goto out_free_state;
637 	}
638 
639 	/*
640 	 * If we read beyond EOD, try unlocking native capacity even if the
641 	 * partition table was successfully read as we could be missing some
642 	 * partitions.
643 	 */
644 	if (state->access_beyond_eod) {
645 		printk(KERN_WARNING
646 		       "%s: partition table partially beyond EOD, ",
647 		       disk->disk_name);
648 		if (disk_unlock_native_capacity(disk))
649 			goto out_free_state;
650 	}
651 
652 	/* tell userspace that the media / partition table may have changed */
653 	kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
654 
655 	for (p = 1; p < state->limit; p++)
656 		if (!blk_add_partition(disk, bdev, state, p))
657 			goto out_free_state;
658 
659 	ret = 0;
660 out_free_state:
661 	free_partitions(state);
662 	return ret;
663 }
664 
665 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
666 {
667 	struct address_space *mapping = state->bdev->bd_inode->i_mapping;
668 	struct page *page;
669 
670 	if (n >= get_capacity(state->bdev->bd_disk)) {
671 		state->access_beyond_eod = true;
672 		return NULL;
673 	}
674 
675 	page = read_mapping_page(mapping,
676 			(pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL);
677 	if (IS_ERR(page))
678 		goto out;
679 	if (PageError(page))
680 		goto out_put_page;
681 
682 	p->v = page;
683 	return (unsigned char *)page_address(page) +
684 			((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT);
685 out_put_page:
686 	put_page(page);
687 out:
688 	p->v = NULL;
689 	return NULL;
690 }
691