xref: /openbmc/linux/block/partitions/core.c (revision dce54888)
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 
15 static int (*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 
88 static 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 
108 static void free_partitions(struct parsed_partitions *state)
109 {
110 	vfree(state->parts);
111 	kfree(state);
112 }
113 
114 static 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 
173 static 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 
179 static 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 
185 static 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 
191 static 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 
197 static 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 
203 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
204 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
205 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
206 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
207 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
208 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
209 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
210 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
211 #ifdef CONFIG_FAIL_MAKE_REQUEST
212 static struct device_attribute dev_attr_fail =
213 	__ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
214 #endif
215 
216 static 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 
231 static struct attribute_group part_attr_group = {
232 	.attrs = part_attrs,
233 };
234 
235 static 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 
243 static 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 
249 static 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 
259 struct device_type part_type = {
260 	.name		= "partition",
261 	.groups		= part_attr_groups,
262 	.release	= part_release,
263 	.uevent		= part_uevent,
264 };
265 
266 static void delete_partition(struct block_device *part)
267 {
268 	lockdep_assert_held(&part->bd_disk->open_mutex);
269 
270 	fsync_bdev(part);
271 	__invalidate_device(part, true);
272 
273 	xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
274 	kobject_put(part->bd_holder_dir);
275 	device_del(&part->bd_device);
276 
277 	/*
278 	 * Remove the block device from the inode hash, so that it cannot be
279 	 * looked up any more even when openers still hold references.
280 	 */
281 	remove_inode_hash(part->bd_inode);
282 
283 	put_device(&part->bd_device);
284 }
285 
286 static ssize_t whole_disk_show(struct device *dev,
287 			       struct device_attribute *attr, char *buf)
288 {
289 	return 0;
290 }
291 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
292 
293 /*
294  * Must be called either with open_mutex held, before a disk can be opened or
295  * after all disk users are gone.
296  */
297 static struct block_device *add_partition(struct gendisk *disk, int partno,
298 				sector_t start, sector_t len, int flags,
299 				struct partition_meta_info *info)
300 {
301 	dev_t devt = MKDEV(0, 0);
302 	struct device *ddev = disk_to_dev(disk);
303 	struct device *pdev;
304 	struct block_device *bdev;
305 	const char *dname;
306 	int err;
307 
308 	lockdep_assert_held(&disk->open_mutex);
309 
310 	if (partno >= DISK_MAX_PARTS)
311 		return ERR_PTR(-EINVAL);
312 
313 	/*
314 	 * Partitions are not supported on zoned block devices that are used as
315 	 * such.
316 	 */
317 	switch (disk->queue->limits.zoned) {
318 	case BLK_ZONED_HM:
319 		pr_warn("%s: partitions not supported on host managed zoned block device\n",
320 			disk->disk_name);
321 		return ERR_PTR(-ENXIO);
322 	case BLK_ZONED_HA:
323 		pr_info("%s: disabling host aware zoned block device support due to partitions\n",
324 			disk->disk_name);
325 		disk_set_zoned(disk, BLK_ZONED_NONE);
326 		break;
327 	case BLK_ZONED_NONE:
328 		break;
329 	}
330 
331 	if (xa_load(&disk->part_tbl, partno))
332 		return ERR_PTR(-EBUSY);
333 
334 	/* ensure we always have a reference to the whole disk */
335 	get_device(disk_to_dev(disk));
336 
337 	err = -ENOMEM;
338 	bdev = bdev_alloc(disk, partno);
339 	if (!bdev)
340 		goto out_put_disk;
341 
342 	bdev->bd_start_sect = start;
343 	bdev_set_nr_sectors(bdev, len);
344 
345 	pdev = &bdev->bd_device;
346 	dname = dev_name(ddev);
347 	if (isdigit(dname[strlen(dname) - 1]))
348 		dev_set_name(pdev, "%sp%d", dname, partno);
349 	else
350 		dev_set_name(pdev, "%s%d", dname, partno);
351 
352 	device_initialize(pdev);
353 	pdev->class = &block_class;
354 	pdev->type = &part_type;
355 	pdev->parent = ddev;
356 
357 	/* in consecutive minor range? */
358 	if (bdev->bd_partno < disk->minors) {
359 		devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno);
360 	} else {
361 		err = blk_alloc_ext_minor();
362 		if (err < 0)
363 			goto out_put;
364 		devt = MKDEV(BLOCK_EXT_MAJOR, err);
365 	}
366 	pdev->devt = devt;
367 
368 	if (info) {
369 		err = -ENOMEM;
370 		bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
371 		if (!bdev->bd_meta_info)
372 			goto out_put;
373 	}
374 
375 	/* delay uevent until 'holders' subdir is created */
376 	dev_set_uevent_suppress(pdev, 1);
377 	err = device_add(pdev);
378 	if (err)
379 		goto out_put;
380 
381 	err = -ENOMEM;
382 	bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
383 	if (!bdev->bd_holder_dir)
384 		goto out_del;
385 
386 	dev_set_uevent_suppress(pdev, 0);
387 	if (flags & ADDPART_FLAG_WHOLEDISK) {
388 		err = device_create_file(pdev, &dev_attr_whole_disk);
389 		if (err)
390 			goto out_del;
391 	}
392 
393 	/* everything is up and running, commence */
394 	err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
395 	if (err)
396 		goto out_del;
397 	bdev_add(bdev, devt);
398 
399 	/* suppress uevent if the disk suppresses it */
400 	if (!dev_get_uevent_suppress(ddev))
401 		kobject_uevent(&pdev->kobj, KOBJ_ADD);
402 	return bdev;
403 
404 out_del:
405 	kobject_put(bdev->bd_holder_dir);
406 	device_del(pdev);
407 out_put:
408 	put_device(pdev);
409 	return ERR_PTR(err);
410 out_put_disk:
411 	put_disk(disk);
412 	return ERR_PTR(err);
413 }
414 
415 static bool partition_overlaps(struct gendisk *disk, sector_t start,
416 		sector_t length, int skip_partno)
417 {
418 	struct block_device *part;
419 	bool overlap = false;
420 	unsigned long idx;
421 
422 	rcu_read_lock();
423 	xa_for_each_start(&disk->part_tbl, idx, part, 1) {
424 		if (part->bd_partno != skip_partno &&
425 		    start < part->bd_start_sect + bdev_nr_sectors(part) &&
426 		    start + length > part->bd_start_sect) {
427 			overlap = true;
428 			break;
429 		}
430 	}
431 	rcu_read_unlock();
432 
433 	return overlap;
434 }
435 
436 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
437 		sector_t length)
438 {
439 	struct block_device *part;
440 	int ret;
441 
442 	mutex_lock(&disk->open_mutex);
443 	if (!disk_live(disk)) {
444 		ret = -ENXIO;
445 		goto out;
446 	}
447 
448 	if (partition_overlaps(disk, start, length, -1)) {
449 		ret = -EBUSY;
450 		goto out;
451 	}
452 
453 	part = add_partition(disk, partno, start, length,
454 			ADDPART_FLAG_NONE, NULL);
455 	ret = PTR_ERR_OR_ZERO(part);
456 out:
457 	mutex_unlock(&disk->open_mutex);
458 	return ret;
459 }
460 
461 int bdev_del_partition(struct gendisk *disk, int partno)
462 {
463 	struct block_device *part = NULL;
464 	int ret = -ENXIO;
465 
466 	mutex_lock(&disk->open_mutex);
467 	part = xa_load(&disk->part_tbl, partno);
468 	if (!part)
469 		goto out_unlock;
470 
471 	ret = -EBUSY;
472 	if (atomic_read(&part->bd_openers))
473 		goto out_unlock;
474 
475 	delete_partition(part);
476 	ret = 0;
477 out_unlock:
478 	mutex_unlock(&disk->open_mutex);
479 	return ret;
480 }
481 
482 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
483 		sector_t length)
484 {
485 	struct block_device *part = NULL;
486 	int ret = -ENXIO;
487 
488 	mutex_lock(&disk->open_mutex);
489 	part = xa_load(&disk->part_tbl, partno);
490 	if (!part)
491 		goto out_unlock;
492 
493 	ret = -EINVAL;
494 	if (start != part->bd_start_sect)
495 		goto out_unlock;
496 
497 	ret = -EBUSY;
498 	if (partition_overlaps(disk, start, length, partno))
499 		goto out_unlock;
500 
501 	bdev_set_nr_sectors(part, length);
502 
503 	ret = 0;
504 out_unlock:
505 	mutex_unlock(&disk->open_mutex);
506 	return ret;
507 }
508 
509 static bool disk_unlock_native_capacity(struct gendisk *disk)
510 {
511 	if (!disk->fops->unlock_native_capacity ||
512 	    test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) {
513 		printk(KERN_CONT "truncated\n");
514 		return false;
515 	}
516 
517 	printk(KERN_CONT "enabling native capacity\n");
518 	disk->fops->unlock_native_capacity(disk);
519 	return true;
520 }
521 
522 void blk_drop_partitions(struct gendisk *disk)
523 {
524 	struct block_device *part;
525 	unsigned long idx;
526 
527 	lockdep_assert_held(&disk->open_mutex);
528 
529 	xa_for_each_start(&disk->part_tbl, idx, part, 1)
530 		delete_partition(part);
531 }
532 
533 static bool blk_add_partition(struct gendisk *disk,
534 		struct parsed_partitions *state, int p)
535 {
536 	sector_t size = state->parts[p].size;
537 	sector_t from = state->parts[p].from;
538 	struct block_device *part;
539 
540 	if (!size)
541 		return true;
542 
543 	if (from >= get_capacity(disk)) {
544 		printk(KERN_WARNING
545 		       "%s: p%d start %llu is beyond EOD, ",
546 		       disk->disk_name, p, (unsigned long long) from);
547 		if (disk_unlock_native_capacity(disk))
548 			return false;
549 		return true;
550 	}
551 
552 	if (from + size > get_capacity(disk)) {
553 		printk(KERN_WARNING
554 		       "%s: p%d size %llu extends beyond EOD, ",
555 		       disk->disk_name, p, (unsigned long long) size);
556 
557 		if (disk_unlock_native_capacity(disk))
558 			return false;
559 
560 		/*
561 		 * We can not ignore partitions of broken tables created by for
562 		 * example camera firmware, but we limit them to the end of the
563 		 * disk to avoid creating invalid block devices.
564 		 */
565 		size = get_capacity(disk) - from;
566 	}
567 
568 	part = add_partition(disk, p, from, size, state->parts[p].flags,
569 			     &state->parts[p].info);
570 	if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
571 		printk(KERN_ERR " %s: p%d could not be added: %ld\n",
572 		       disk->disk_name, p, -PTR_ERR(part));
573 		return true;
574 	}
575 
576 	if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
577 	    (state->parts[p].flags & ADDPART_FLAG_RAID))
578 		md_autodetect_dev(part->bd_dev);
579 
580 	return true;
581 }
582 
583 static int blk_add_partitions(struct gendisk *disk)
584 {
585 	struct parsed_partitions *state;
586 	int ret = -EAGAIN, p;
587 
588 	if (disk->flags & GENHD_FL_NO_PART)
589 		return 0;
590 
591 	if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
592 		return 0;
593 
594 	state = check_partition(disk);
595 	if (!state)
596 		return 0;
597 	if (IS_ERR(state)) {
598 		/*
599 		 * I/O error reading the partition table.  If we tried to read
600 		 * beyond EOD, retry after unlocking the native capacity.
601 		 */
602 		if (PTR_ERR(state) == -ENOSPC) {
603 			printk(KERN_WARNING "%s: partition table beyond EOD, ",
604 			       disk->disk_name);
605 			if (disk_unlock_native_capacity(disk))
606 				return -EAGAIN;
607 		}
608 		return -EIO;
609 	}
610 
611 	/*
612 	 * Partitions are not supported on host managed zoned block devices.
613 	 */
614 	if (disk->queue->limits.zoned == BLK_ZONED_HM) {
615 		pr_warn("%s: ignoring partition table on host managed zoned block device\n",
616 			disk->disk_name);
617 		ret = 0;
618 		goto out_free_state;
619 	}
620 
621 	/*
622 	 * If we read beyond EOD, try unlocking native capacity even if the
623 	 * partition table was successfully read as we could be missing some
624 	 * partitions.
625 	 */
626 	if (state->access_beyond_eod) {
627 		printk(KERN_WARNING
628 		       "%s: partition table partially beyond EOD, ",
629 		       disk->disk_name);
630 		if (disk_unlock_native_capacity(disk))
631 			goto out_free_state;
632 	}
633 
634 	/* tell userspace that the media / partition table may have changed */
635 	kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
636 
637 	for (p = 1; p < state->limit; p++)
638 		if (!blk_add_partition(disk, state, p))
639 			goto out_free_state;
640 
641 	ret = 0;
642 out_free_state:
643 	free_partitions(state);
644 	return ret;
645 }
646 
647 int bdev_disk_changed(struct gendisk *disk, bool invalidate)
648 {
649 	int ret = 0;
650 
651 	lockdep_assert_held(&disk->open_mutex);
652 
653 	if (!disk_live(disk))
654 		return -ENXIO;
655 
656 rescan:
657 	if (disk->open_partitions)
658 		return -EBUSY;
659 	sync_blockdev(disk->part0);
660 	invalidate_bdev(disk->part0);
661 	blk_drop_partitions(disk);
662 
663 	clear_bit(GD_NEED_PART_SCAN, &disk->state);
664 
665 	/*
666 	 * Historically we only set the capacity to zero for devices that
667 	 * support partitions (independ of actually having partitions created).
668 	 * Doing that is rather inconsistent, but changing it broke legacy
669 	 * udisks polling for legacy ide-cdrom devices.  Use the crude check
670 	 * below to get the sane behavior for most device while not breaking
671 	 * userspace for this particular setup.
672 	 */
673 	if (invalidate) {
674 		if (!(disk->flags & GENHD_FL_NO_PART) ||
675 		    !(disk->flags & GENHD_FL_REMOVABLE))
676 			set_capacity(disk, 0);
677 	}
678 
679 	if (get_capacity(disk)) {
680 		ret = blk_add_partitions(disk);
681 		if (ret == -EAGAIN)
682 			goto rescan;
683 	} else if (invalidate) {
684 		/*
685 		 * Tell userspace that the media / partition table may have
686 		 * changed.
687 		 */
688 		kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
689 	}
690 
691 	return ret;
692 }
693 /*
694  * Only exported for loop and dasd for historic reasons.  Don't use in new
695  * code!
696  */
697 EXPORT_SYMBOL_GPL(bdev_disk_changed);
698 
699 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
700 {
701 	struct address_space *mapping = state->disk->part0->bd_inode->i_mapping;
702 	struct folio *folio;
703 
704 	if (n >= get_capacity(state->disk)) {
705 		state->access_beyond_eod = true;
706 		goto out;
707 	}
708 
709 	folio = read_mapping_folio(mapping, n >> PAGE_SECTORS_SHIFT, NULL);
710 	if (IS_ERR(folio))
711 		goto out;
712 
713 	p->v = folio;
714 	return folio_address(folio) + offset_in_folio(folio, n * SECTOR_SIZE);
715 out:
716 	p->v = NULL;
717 	return NULL;
718 }
719