xref: /openbmc/linux/block/partitions/core.c (revision 6f4eaea2)
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 	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);
94 }
95 
96 static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
97 {
98 	struct parsed_partitions *state;
99 	int nr;
100 
101 	state = kzalloc(sizeof(*state), GFP_KERNEL);
102 	if (!state)
103 		return NULL;
104 
105 	nr = disk_max_parts(hd);
106 	state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
107 	if (!state->parts) {
108 		kfree(state);
109 		return NULL;
110 	}
111 
112 	state->limit = nr;
113 
114 	return state;
115 }
116 
117 static void free_partitions(struct parsed_partitions *state)
118 {
119 	vfree(state->parts);
120 	kfree(state);
121 }
122 
123 static struct parsed_partitions *check_partition(struct gendisk *hd,
124 		struct block_device *bdev)
125 {
126 	struct parsed_partitions *state;
127 	int i, res, err;
128 
129 	state = allocate_partitions(hd);
130 	if (!state)
131 		return NULL;
132 	state->pp_buf = (char *)__get_free_page(GFP_KERNEL);
133 	if (!state->pp_buf) {
134 		free_partitions(state);
135 		return NULL;
136 	}
137 	state->pp_buf[0] = '\0';
138 
139 	state->bdev = bdev;
140 	disk_name(hd, 0, state->name);
141 	snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name);
142 	if (isdigit(state->name[strlen(state->name)-1]))
143 		sprintf(state->name, "p");
144 
145 	i = res = err = 0;
146 	while (!res && check_part[i]) {
147 		memset(state->parts, 0, state->limit * sizeof(state->parts[0]));
148 		res = check_part[i++](state);
149 		if (res < 0) {
150 			/*
151 			 * We have hit an I/O error which we don't report now.
152 			 * But record it, and let the others do their job.
153 			 */
154 			err = res;
155 			res = 0;
156 		}
157 
158 	}
159 	if (res > 0) {
160 		printk(KERN_INFO "%s", state->pp_buf);
161 
162 		free_page((unsigned long)state->pp_buf);
163 		return state;
164 	}
165 	if (state->access_beyond_eod)
166 		err = -ENOSPC;
167 	/*
168 	 * The partition is unrecognized. So report I/O errors if there were any
169 	 */
170 	if (err)
171 		res = err;
172 	if (res) {
173 		strlcat(state->pp_buf,
174 			" unable to read partition table\n", PAGE_SIZE);
175 		printk(KERN_INFO "%s", state->pp_buf);
176 	}
177 
178 	free_page((unsigned long)state->pp_buf);
179 	free_partitions(state);
180 	return ERR_PTR(res);
181 }
182 
183 static ssize_t part_partition_show(struct device *dev,
184 				   struct device_attribute *attr, char *buf)
185 {
186 	return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno);
187 }
188 
189 static ssize_t part_start_show(struct device *dev,
190 			       struct device_attribute *attr, char *buf)
191 {
192 	return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect);
193 }
194 
195 static ssize_t part_ro_show(struct device *dev,
196 			    struct device_attribute *attr, char *buf)
197 {
198 	return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev)));
199 }
200 
201 static ssize_t part_alignment_offset_show(struct device *dev,
202 					  struct device_attribute *attr, char *buf)
203 {
204 	struct block_device *bdev = dev_to_bdev(dev);
205 
206 	return sprintf(buf, "%u\n",
207 		queue_limit_alignment_offset(&bdev->bd_disk->queue->limits,
208 				bdev->bd_start_sect));
209 }
210 
211 static ssize_t part_discard_alignment_show(struct device *dev,
212 					   struct device_attribute *attr, char *buf)
213 {
214 	struct block_device *bdev = dev_to_bdev(dev);
215 
216 	return sprintf(buf, "%u\n",
217 		queue_limit_discard_alignment(&bdev->bd_disk->queue->limits,
218 				bdev->bd_start_sect));
219 }
220 
221 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
222 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
223 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
224 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
225 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
226 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
227 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
228 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
229 #ifdef CONFIG_FAIL_MAKE_REQUEST
230 static struct device_attribute dev_attr_fail =
231 	__ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
232 #endif
233 
234 static struct attribute *part_attrs[] = {
235 	&dev_attr_partition.attr,
236 	&dev_attr_start.attr,
237 	&dev_attr_size.attr,
238 	&dev_attr_ro.attr,
239 	&dev_attr_alignment_offset.attr,
240 	&dev_attr_discard_alignment.attr,
241 	&dev_attr_stat.attr,
242 	&dev_attr_inflight.attr,
243 #ifdef CONFIG_FAIL_MAKE_REQUEST
244 	&dev_attr_fail.attr,
245 #endif
246 	NULL
247 };
248 
249 static struct attribute_group part_attr_group = {
250 	.attrs = part_attrs,
251 };
252 
253 static const struct attribute_group *part_attr_groups[] = {
254 	&part_attr_group,
255 #ifdef CONFIG_BLK_DEV_IO_TRACE
256 	&blk_trace_attr_group,
257 #endif
258 	NULL
259 };
260 
261 static void part_release(struct device *dev)
262 {
263 	blk_free_devt(dev->devt);
264 	bdput(dev_to_bdev(dev));
265 }
266 
267 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
268 {
269 	struct block_device *part = dev_to_bdev(dev);
270 
271 	add_uevent_var(env, "PARTN=%u", part->bd_partno);
272 	if (part->bd_meta_info && part->bd_meta_info->volname[0])
273 		add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname);
274 	return 0;
275 }
276 
277 struct device_type part_type = {
278 	.name		= "partition",
279 	.groups		= part_attr_groups,
280 	.release	= part_release,
281 	.uevent		= part_uevent,
282 };
283 
284 /*
285  * Must be called either with bd_mutex held, before a disk can be opened or
286  * after all disk users are gone.
287  */
288 void delete_partition(struct block_device *part)
289 {
290 	xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
291 	kobject_put(part->bd_holder_dir);
292 	device_del(&part->bd_device);
293 
294 	/*
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.
297 	 */
298 	remove_inode_hash(part->bd_inode);
299 
300 	put_device(&part->bd_device);
301 }
302 
303 static ssize_t whole_disk_show(struct device *dev,
304 			       struct device_attribute *attr, char *buf)
305 {
306 	return 0;
307 }
308 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
309 
310 /*
311  * Must be called either with bd_mutex held, before a disk can be opened or
312  * after all disk users are gone.
313  */
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)
317 {
318 	dev_t devt = MKDEV(0, 0);
319 	struct device *ddev = disk_to_dev(disk);
320 	struct device *pdev;
321 	struct block_device *bdev;
322 	const char *dname;
323 	int err;
324 
325 	/*
326 	 * Partitions are not supported on zoned block devices that are used as
327 	 * such.
328 	 */
329 	switch (disk->queue->limits.zoned) {
330 	case BLK_ZONED_HM:
331 		pr_warn("%s: partitions not supported on host managed zoned block device\n",
332 			disk->disk_name);
333 		return ERR_PTR(-ENXIO);
334 	case BLK_ZONED_HA:
335 		pr_info("%s: disabling host aware zoned block device support due to partitions\n",
336 			disk->disk_name);
337 		blk_queue_set_zoned(disk, BLK_ZONED_NONE);
338 		break;
339 	case BLK_ZONED_NONE:
340 		break;
341 	}
342 
343 	if (xa_load(&disk->part_tbl, partno))
344 		return ERR_PTR(-EBUSY);
345 
346 	bdev = bdev_alloc(disk, partno);
347 	if (!bdev)
348 		return ERR_PTR(-ENOMEM);
349 
350 	bdev->bd_start_sect = start;
351 	bdev_set_nr_sectors(bdev, len);
352 
353 	if (info) {
354 		err = -ENOMEM;
355 		bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
356 		if (!bdev->bd_meta_info)
357 			goto out_bdput;
358 	}
359 
360 	pdev = &bdev->bd_device;
361 	dname = dev_name(ddev);
362 	if (isdigit(dname[strlen(dname) - 1]))
363 		dev_set_name(pdev, "%sp%d", dname, partno);
364 	else
365 		dev_set_name(pdev, "%s%d", dname, partno);
366 
367 	device_initialize(pdev);
368 	pdev->class = &block_class;
369 	pdev->type = &part_type;
370 	pdev->parent = ddev;
371 
372 	err = blk_alloc_devt(bdev, &devt);
373 	if (err)
374 		goto out_put;
375 	pdev->devt = devt;
376 
377 	/* delay uevent until 'holders' subdir is created */
378 	dev_set_uevent_suppress(pdev, 1);
379 	err = device_add(pdev);
380 	if (err)
381 		goto out_put;
382 
383 	err = -ENOMEM;
384 	bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
385 	if (!bdev->bd_holder_dir)
386 		goto out_del;
387 
388 	dev_set_uevent_suppress(pdev, 0);
389 	if (flags & ADDPART_FLAG_WHOLEDISK) {
390 		err = device_create_file(pdev, &dev_attr_whole_disk);
391 		if (err)
392 			goto out_del;
393 	}
394 
395 	/* everything is up and running, commence */
396 	err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
397 	if (err)
398 		goto out_del;
399 	bdev_add(bdev, devt);
400 
401 	/* suppress uevent if the disk suppresses it */
402 	if (!dev_get_uevent_suppress(ddev))
403 		kobject_uevent(&pdev->kobj, KOBJ_ADD);
404 	return bdev;
405 
406 out_bdput:
407 	bdput(bdev);
408 	return ERR_PTR(err);
409 out_del:
410 	kobject_put(bdev->bd_holder_dir);
411 	device_del(pdev);
412 out_put:
413 	put_device(pdev);
414 	return ERR_PTR(err);
415 }
416 
417 static bool partition_overlaps(struct gendisk *disk, sector_t start,
418 		sector_t length, int skip_partno)
419 {
420 	struct disk_part_iter piter;
421 	struct block_device *part;
422 	bool overlap = false;
423 
424 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
425 	while ((part = disk_part_iter_next(&piter))) {
426 		if (part->bd_partno == skip_partno ||
427 		    start >= part->bd_start_sect + bdev_nr_sectors(part) ||
428 		    start + length <= part->bd_start_sect)
429 			continue;
430 		overlap = true;
431 		break;
432 	}
433 
434 	disk_part_iter_exit(&piter);
435 	return overlap;
436 }
437 
438 int bdev_add_partition(struct block_device *bdev, int partno,
439 		sector_t start, sector_t length)
440 {
441 	struct block_device *part;
442 
443 	mutex_lock(&bdev->bd_mutex);
444 	if (partition_overlaps(bdev->bd_disk, start, length, -1)) {
445 		mutex_unlock(&bdev->bd_mutex);
446 		return -EBUSY;
447 	}
448 
449 	part = add_partition(bdev->bd_disk, partno, start, length,
450 			ADDPART_FLAG_NONE, NULL);
451 	mutex_unlock(&bdev->bd_mutex);
452 	return PTR_ERR_OR_ZERO(part);
453 }
454 
455 int bdev_del_partition(struct block_device *bdev, int partno)
456 {
457 	struct block_device *part;
458 	int ret;
459 
460 	part = bdget_disk(bdev->bd_disk, partno);
461 	if (!part)
462 		return -ENXIO;
463 
464 	mutex_lock(&part->bd_mutex);
465 	mutex_lock_nested(&bdev->bd_mutex, 1);
466 
467 	ret = -EBUSY;
468 	if (part->bd_openers)
469 		goto out_unlock;
470 
471 	sync_blockdev(part);
472 	invalidate_bdev(part);
473 
474 	delete_partition(part);
475 	ret = 0;
476 out_unlock:
477 	mutex_unlock(&bdev->bd_mutex);
478 	mutex_unlock(&part->bd_mutex);
479 	bdput(part);
480 	return ret;
481 }
482 
483 int bdev_resize_partition(struct block_device *bdev, int partno,
484 		sector_t start, sector_t length)
485 {
486 	struct block_device *part;
487 	int ret = 0;
488 
489 	part = bdget_disk(bdev->bd_disk, partno);
490 	if (!part)
491 		return -ENXIO;
492 
493 	mutex_lock(&part->bd_mutex);
494 	mutex_lock_nested(&bdev->bd_mutex, 1);
495 	ret = -EINVAL;
496 	if (start != part->bd_start_sect)
497 		goto out_unlock;
498 
499 	ret = -EBUSY;
500 	if (partition_overlaps(bdev->bd_disk, start, length, partno))
501 		goto out_unlock;
502 
503 	bdev_set_nr_sectors(part, length);
504 
505 	ret = 0;
506 out_unlock:
507 	mutex_unlock(&part->bd_mutex);
508 	mutex_unlock(&bdev->bd_mutex);
509 	bdput(part);
510 	return ret;
511 }
512 
513 static bool disk_unlock_native_capacity(struct gendisk *disk)
514 {
515 	const struct block_device_operations *bdops = disk->fops;
516 
517 	if (bdops->unlock_native_capacity &&
518 	    !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
519 		printk(KERN_CONT "enabling native capacity\n");
520 		bdops->unlock_native_capacity(disk);
521 		disk->flags |= GENHD_FL_NATIVE_CAPACITY;
522 		return true;
523 	} else {
524 		printk(KERN_CONT "truncated\n");
525 		return false;
526 	}
527 }
528 
529 int blk_drop_partitions(struct block_device *bdev)
530 {
531 	struct disk_part_iter piter;
532 	struct block_device *part;
533 
534 	if (bdev->bd_part_count)
535 		return -EBUSY;
536 
537 	sync_blockdev(bdev);
538 	invalidate_bdev(bdev);
539 
540 	disk_part_iter_init(&piter, bdev->bd_disk, DISK_PITER_INCL_EMPTY);
541 	while ((part = disk_part_iter_next(&piter)))
542 		delete_partition(part);
543 	disk_part_iter_exit(&piter);
544 
545 	return 0;
546 }
547 #ifdef CONFIG_S390
548 /* for historic reasons in the DASD driver */
549 EXPORT_SYMBOL_GPL(blk_drop_partitions);
550 #endif
551 
552 static bool blk_add_partition(struct gendisk *disk, struct block_device *bdev,
553 		struct parsed_partitions *state, int p)
554 {
555 	sector_t size = state->parts[p].size;
556 	sector_t from = state->parts[p].from;
557 	struct block_device *part;
558 
559 	if (!size)
560 		return true;
561 
562 	if (from >= get_capacity(disk)) {
563 		printk(KERN_WARNING
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))
567 			return false;
568 		return true;
569 	}
570 
571 	if (from + size > get_capacity(disk)) {
572 		printk(KERN_WARNING
573 		       "%s: p%d size %llu extends beyond EOD, ",
574 		       disk->disk_name, p, (unsigned long long) size);
575 
576 		if (disk_unlock_native_capacity(disk))
577 			return false;
578 
579 		/*
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.
583 		 */
584 		size = get_capacity(disk) - from;
585 	}
586 
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));
592 		return true;
593 	}
594 
595 	if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
596 	    (state->parts[p].flags & ADDPART_FLAG_RAID))
597 		md_autodetect_dev(part->bd_dev);
598 
599 	return true;
600 }
601 
602 int blk_add_partitions(struct gendisk *disk, struct block_device *bdev)
603 {
604 	struct parsed_partitions *state;
605 	int ret = -EAGAIN, p;
606 
607 	if (!disk_part_scan_enabled(disk))
608 		return 0;
609 
610 	state = check_partition(disk, bdev);
611 	if (!state)
612 		return 0;
613 	if (IS_ERR(state)) {
614 		/*
615 		 * I/O error reading the partition table.  If we tried to read
616 		 * beyond EOD, retry after unlocking the native capacity.
617 		 */
618 		if (PTR_ERR(state) == -ENOSPC) {
619 			printk(KERN_WARNING "%s: partition table beyond EOD, ",
620 			       disk->disk_name);
621 			if (disk_unlock_native_capacity(disk))
622 				return -EAGAIN;
623 		}
624 		return -EIO;
625 	}
626 
627 	/*
628 	 * Partitions are not supported on host managed zoned block devices.
629 	 */
630 	if (disk->queue->limits.zoned == BLK_ZONED_HM) {
631 		pr_warn("%s: ignoring partition table on host managed zoned block device\n",
632 			disk->disk_name);
633 		ret = 0;
634 		goto out_free_state;
635 	}
636 
637 	/*
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
640 	 * partitions.
641 	 */
642 	if (state->access_beyond_eod) {
643 		printk(KERN_WARNING
644 		       "%s: partition table partially beyond EOD, ",
645 		       disk->disk_name);
646 		if (disk_unlock_native_capacity(disk))
647 			goto out_free_state;
648 	}
649 
650 	/* tell userspace that the media / partition table may have changed */
651 	kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
652 
653 	for (p = 1; p < state->limit; p++)
654 		if (!blk_add_partition(disk, bdev, state, p))
655 			goto out_free_state;
656 
657 	ret = 0;
658 out_free_state:
659 	free_partitions(state);
660 	return ret;
661 }
662 
663 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
664 {
665 	struct address_space *mapping = state->bdev->bd_inode->i_mapping;
666 	struct page *page;
667 
668 	if (n >= get_capacity(state->bdev->bd_disk)) {
669 		state->access_beyond_eod = true;
670 		return NULL;
671 	}
672 
673 	page = read_mapping_page(mapping,
674 			(pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL);
675 	if (IS_ERR(page))
676 		goto out;
677 	if (PageError(page))
678 		goto out_put_page;
679 
680 	p->v = page;
681 	return (unsigned char *)page_address(page) +
682 			((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT);
683 out_put_page:
684 	put_page(page);
685 out:
686 	p->v = NULL;
687 	return NULL;
688 }
689