xref: /openbmc/linux/block/partitions/core.c (revision 877013bc)
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 static void delete_partition(struct block_device *part)
289 {
290 	fsync_bdev(part);
291 	__invalidate_device(part, true);
292 
293 	xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
294 	kobject_put(part->bd_holder_dir);
295 	device_del(&part->bd_device);
296 
297 	/*
298 	 * Remove the block device from the inode hash, so that it cannot be
299 	 * looked up any more even when openers still hold references.
300 	 */
301 	remove_inode_hash(part->bd_inode);
302 
303 	put_device(&part->bd_device);
304 }
305 
306 static ssize_t whole_disk_show(struct device *dev,
307 			       struct device_attribute *attr, char *buf)
308 {
309 	return 0;
310 }
311 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
312 
313 /*
314  * Must be called either with bd_mutex held, before a disk can be opened or
315  * after all disk users are gone.
316  */
317 static struct block_device *add_partition(struct gendisk *disk, int partno,
318 				sector_t start, sector_t len, int flags,
319 				struct partition_meta_info *info)
320 {
321 	dev_t devt = MKDEV(0, 0);
322 	struct device *ddev = disk_to_dev(disk);
323 	struct device *pdev;
324 	struct block_device *bdev;
325 	const char *dname;
326 	int err;
327 
328 	/*
329 	 * disk_max_parts() won't be zero, either GENHD_FL_EXT_DEVT is set
330 	 * or 'minors' is passed to alloc_disk().
331 	 */
332 	if (partno >= disk_max_parts(disk))
333 		return ERR_PTR(-EINVAL);
334 
335 	/*
336 	 * Partitions are not supported on zoned block devices that are used as
337 	 * such.
338 	 */
339 	switch (disk->queue->limits.zoned) {
340 	case BLK_ZONED_HM:
341 		pr_warn("%s: partitions not supported on host managed zoned block device\n",
342 			disk->disk_name);
343 		return ERR_PTR(-ENXIO);
344 	case BLK_ZONED_HA:
345 		pr_info("%s: disabling host aware zoned block device support due to partitions\n",
346 			disk->disk_name);
347 		blk_queue_set_zoned(disk, BLK_ZONED_NONE);
348 		break;
349 	case BLK_ZONED_NONE:
350 		break;
351 	}
352 
353 	if (xa_load(&disk->part_tbl, partno))
354 		return ERR_PTR(-EBUSY);
355 
356 	bdev = bdev_alloc(disk, partno);
357 	if (!bdev)
358 		return ERR_PTR(-ENOMEM);
359 
360 	bdev->bd_start_sect = start;
361 	bdev_set_nr_sectors(bdev, len);
362 
363 	if (info) {
364 		err = -ENOMEM;
365 		bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
366 		if (!bdev->bd_meta_info)
367 			goto out_bdput;
368 	}
369 
370 	pdev = &bdev->bd_device;
371 	dname = dev_name(ddev);
372 	if (isdigit(dname[strlen(dname) - 1]))
373 		dev_set_name(pdev, "%sp%d", dname, partno);
374 	else
375 		dev_set_name(pdev, "%s%d", dname, partno);
376 
377 	device_initialize(pdev);
378 	pdev->class = &block_class;
379 	pdev->type = &part_type;
380 	pdev->parent = ddev;
381 
382 	err = blk_alloc_devt(bdev, &devt);
383 	if (err)
384 		goto out_put;
385 	pdev->devt = devt;
386 
387 	/* delay uevent until 'holders' subdir is created */
388 	dev_set_uevent_suppress(pdev, 1);
389 	err = device_add(pdev);
390 	if (err)
391 		goto out_put;
392 
393 	err = -ENOMEM;
394 	bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
395 	if (!bdev->bd_holder_dir)
396 		goto out_del;
397 
398 	dev_set_uevent_suppress(pdev, 0);
399 	if (flags & ADDPART_FLAG_WHOLEDISK) {
400 		err = device_create_file(pdev, &dev_attr_whole_disk);
401 		if (err)
402 			goto out_del;
403 	}
404 
405 	/* everything is up and running, commence */
406 	err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
407 	if (err)
408 		goto out_del;
409 	bdev_add(bdev, devt);
410 
411 	/* suppress uevent if the disk suppresses it */
412 	if (!dev_get_uevent_suppress(ddev))
413 		kobject_uevent(&pdev->kobj, KOBJ_ADD);
414 	return bdev;
415 
416 out_bdput:
417 	bdput(bdev);
418 	return ERR_PTR(err);
419 out_del:
420 	kobject_put(bdev->bd_holder_dir);
421 	device_del(pdev);
422 out_put:
423 	put_device(pdev);
424 	return ERR_PTR(err);
425 }
426 
427 static bool partition_overlaps(struct gendisk *disk, sector_t start,
428 		sector_t length, int skip_partno)
429 {
430 	struct block_device *part;
431 	bool overlap = false;
432 	unsigned long idx;
433 
434 	rcu_read_lock();
435 	xa_for_each_start(&disk->part_tbl, idx, part, 1) {
436 		if (part->bd_partno != skip_partno &&
437 		    start < part->bd_start_sect + bdev_nr_sectors(part) &&
438 		    start + length > part->bd_start_sect) {
439 			overlap = true;
440 			break;
441 		}
442 	}
443 	rcu_read_unlock();
444 
445 	return overlap;
446 }
447 
448 int bdev_add_partition(struct block_device *bdev, int partno,
449 		sector_t start, sector_t length)
450 {
451 	struct block_device *part;
452 
453 	mutex_lock(&bdev->bd_mutex);
454 	if (partition_overlaps(bdev->bd_disk, start, length, -1)) {
455 		mutex_unlock(&bdev->bd_mutex);
456 		return -EBUSY;
457 	}
458 
459 	part = add_partition(bdev->bd_disk, partno, start, length,
460 			ADDPART_FLAG_NONE, NULL);
461 	mutex_unlock(&bdev->bd_mutex);
462 	return PTR_ERR_OR_ZERO(part);
463 }
464 
465 int bdev_del_partition(struct block_device *bdev, int partno)
466 {
467 	struct block_device *part;
468 	int ret;
469 
470 	part = bdget_disk(bdev->bd_disk, partno);
471 	if (!part)
472 		return -ENXIO;
473 
474 	mutex_lock(&part->bd_mutex);
475 	mutex_lock_nested(&bdev->bd_mutex, 1);
476 
477 	ret = -EBUSY;
478 	if (part->bd_openers)
479 		goto out_unlock;
480 
481 	delete_partition(part);
482 	ret = 0;
483 out_unlock:
484 	mutex_unlock(&bdev->bd_mutex);
485 	mutex_unlock(&part->bd_mutex);
486 	bdput(part);
487 	return ret;
488 }
489 
490 int bdev_resize_partition(struct block_device *bdev, int partno,
491 		sector_t start, sector_t length)
492 {
493 	struct block_device *part;
494 	int ret = 0;
495 
496 	part = bdget_disk(bdev->bd_disk, partno);
497 	if (!part)
498 		return -ENXIO;
499 
500 	mutex_lock(&part->bd_mutex);
501 	mutex_lock_nested(&bdev->bd_mutex, 1);
502 	ret = -EINVAL;
503 	if (start != part->bd_start_sect)
504 		goto out_unlock;
505 
506 	ret = -EBUSY;
507 	if (partition_overlaps(bdev->bd_disk, start, length, partno))
508 		goto out_unlock;
509 
510 	bdev_set_nr_sectors(part, length);
511 
512 	ret = 0;
513 out_unlock:
514 	mutex_unlock(&part->bd_mutex);
515 	mutex_unlock(&bdev->bd_mutex);
516 	bdput(part);
517 	return ret;
518 }
519 
520 static bool disk_unlock_native_capacity(struct gendisk *disk)
521 {
522 	const struct block_device_operations *bdops = disk->fops;
523 
524 	if (bdops->unlock_native_capacity &&
525 	    !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
526 		printk(KERN_CONT "enabling native capacity\n");
527 		bdops->unlock_native_capacity(disk);
528 		disk->flags |= GENHD_FL_NATIVE_CAPACITY;
529 		return true;
530 	} else {
531 		printk(KERN_CONT "truncated\n");
532 		return false;
533 	}
534 }
535 
536 void blk_drop_partitions(struct gendisk *disk)
537 {
538 	struct block_device *part;
539 	unsigned long idx;
540 
541 	lockdep_assert_held(&disk->part0->bd_mutex);
542 
543 	xa_for_each_start(&disk->part_tbl, idx, part, 1) {
544 		if (!bdgrab(part))
545 			continue;
546 		delete_partition(part);
547 		bdput(part);
548 	}
549 }
550 
551 static bool blk_add_partition(struct gendisk *disk, struct block_device *bdev,
552 		struct parsed_partitions *state, int p)
553 {
554 	sector_t size = state->parts[p].size;
555 	sector_t from = state->parts[p].from;
556 	struct block_device *part;
557 
558 	if (!size)
559 		return true;
560 
561 	if (from >= get_capacity(disk)) {
562 		printk(KERN_WARNING
563 		       "%s: p%d start %llu is beyond EOD, ",
564 		       disk->disk_name, p, (unsigned long long) from);
565 		if (disk_unlock_native_capacity(disk))
566 			return false;
567 		return true;
568 	}
569 
570 	if (from + size > get_capacity(disk)) {
571 		printk(KERN_WARNING
572 		       "%s: p%d size %llu extends beyond EOD, ",
573 		       disk->disk_name, p, (unsigned long long) size);
574 
575 		if (disk_unlock_native_capacity(disk))
576 			return false;
577 
578 		/*
579 		 * We can not ignore partitions of broken tables created by for
580 		 * example camera firmware, but we limit them to the end of the
581 		 * disk to avoid creating invalid block devices.
582 		 */
583 		size = get_capacity(disk) - from;
584 	}
585 
586 	part = add_partition(disk, p, from, size, state->parts[p].flags,
587 			     &state->parts[p].info);
588 	if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
589 		printk(KERN_ERR " %s: p%d could not be added: %ld\n",
590 		       disk->disk_name, p, -PTR_ERR(part));
591 		return true;
592 	}
593 
594 	if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
595 	    (state->parts[p].flags & ADDPART_FLAG_RAID))
596 		md_autodetect_dev(part->bd_dev);
597 
598 	return true;
599 }
600 
601 int blk_add_partitions(struct gendisk *disk, struct block_device *bdev)
602 {
603 	struct parsed_partitions *state;
604 	int ret = -EAGAIN, p;
605 
606 	if (!disk_part_scan_enabled(disk))
607 		return 0;
608 
609 	state = check_partition(disk, bdev);
610 	if (!state)
611 		return 0;
612 	if (IS_ERR(state)) {
613 		/*
614 		 * I/O error reading the partition table.  If we tried to read
615 		 * beyond EOD, retry after unlocking the native capacity.
616 		 */
617 		if (PTR_ERR(state) == -ENOSPC) {
618 			printk(KERN_WARNING "%s: partition table beyond EOD, ",
619 			       disk->disk_name);
620 			if (disk_unlock_native_capacity(disk))
621 				return -EAGAIN;
622 		}
623 		return -EIO;
624 	}
625 
626 	/*
627 	 * Partitions are not supported on host managed zoned block devices.
628 	 */
629 	if (disk->queue->limits.zoned == BLK_ZONED_HM) {
630 		pr_warn("%s: ignoring partition table on host managed zoned block device\n",
631 			disk->disk_name);
632 		ret = 0;
633 		goto out_free_state;
634 	}
635 
636 	/*
637 	 * If we read beyond EOD, try unlocking native capacity even if the
638 	 * partition table was successfully read as we could be missing some
639 	 * partitions.
640 	 */
641 	if (state->access_beyond_eod) {
642 		printk(KERN_WARNING
643 		       "%s: partition table partially beyond EOD, ",
644 		       disk->disk_name);
645 		if (disk_unlock_native_capacity(disk))
646 			goto out_free_state;
647 	}
648 
649 	/* tell userspace that the media / partition table may have changed */
650 	kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
651 
652 	for (p = 1; p < state->limit; p++)
653 		if (!blk_add_partition(disk, bdev, state, p))
654 			goto out_free_state;
655 
656 	ret = 0;
657 out_free_state:
658 	free_partitions(state);
659 	return ret;
660 }
661 
662 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
663 {
664 	struct address_space *mapping = state->bdev->bd_inode->i_mapping;
665 	struct page *page;
666 
667 	if (n >= get_capacity(state->bdev->bd_disk)) {
668 		state->access_beyond_eod = true;
669 		return NULL;
670 	}
671 
672 	page = read_mapping_page(mapping,
673 			(pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL);
674 	if (IS_ERR(page))
675 		goto out;
676 	if (PageError(page))
677 		goto out_put_page;
678 
679 	p->v = page;
680 	return (unsigned char *)page_address(page) +
681 			((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT);
682 out_put_page:
683 	put_page(page);
684 out:
685 	p->v = NULL;
686 	return NULL;
687 }
688