]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - block/partitions/core.c
b9e9af84f5188d53b2688cf09f3b35a0f6cae5c0
[mirror_ubuntu-jammy-kernel.git] / block / partitions / core.c
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/genhd.h>
12 #include <linux/vmalloc.h>
13 #include <linux/blktrace_api.h>
14 #include <linux/raid/detect.h>
15 #include "check.h"
16
17 static int (*check_part[])(struct parsed_partitions *) = {
18 /*
19 * Probe partition formats with tables at disk address 0
20 * that also have an ADFS boot block at 0xdc0.
21 */
22 #ifdef CONFIG_ACORN_PARTITION_ICS
23 adfspart_check_ICS,
24 #endif
25 #ifdef CONFIG_ACORN_PARTITION_POWERTEC
26 adfspart_check_POWERTEC,
27 #endif
28 #ifdef CONFIG_ACORN_PARTITION_EESOX
29 adfspart_check_EESOX,
30 #endif
31
32 /*
33 * Now move on to formats that only have partition info at
34 * disk address 0xdc0. Since these may also have stale
35 * PC/BIOS partition tables, they need to come before
36 * the msdos entry.
37 */
38 #ifdef CONFIG_ACORN_PARTITION_CUMANA
39 adfspart_check_CUMANA,
40 #endif
41 #ifdef CONFIG_ACORN_PARTITION_ADFS
42 adfspart_check_ADFS,
43 #endif
44
45 #ifdef CONFIG_CMDLINE_PARTITION
46 cmdline_partition,
47 #endif
48 #ifdef CONFIG_EFI_PARTITION
49 efi_partition, /* this must come before msdos */
50 #endif
51 #ifdef CONFIG_SGI_PARTITION
52 sgi_partition,
53 #endif
54 #ifdef CONFIG_LDM_PARTITION
55 ldm_partition, /* this must come before msdos */
56 #endif
57 #ifdef CONFIG_MSDOS_PARTITION
58 msdos_partition,
59 #endif
60 #ifdef CONFIG_OSF_PARTITION
61 osf_partition,
62 #endif
63 #ifdef CONFIG_SUN_PARTITION
64 sun_partition,
65 #endif
66 #ifdef CONFIG_AMIGA_PARTITION
67 amiga_partition,
68 #endif
69 #ifdef CONFIG_ATARI_PARTITION
70 atari_partition,
71 #endif
72 #ifdef CONFIG_MAC_PARTITION
73 mac_partition,
74 #endif
75 #ifdef CONFIG_ULTRIX_PARTITION
76 ultrix_partition,
77 #endif
78 #ifdef CONFIG_IBM_PARTITION
79 ibm_partition,
80 #endif
81 #ifdef CONFIG_KARMA_PARTITION
82 karma_partition,
83 #endif
84 #ifdef CONFIG_SYSV68_PARTITION
85 sysv68_partition,
86 #endif
87 NULL
88 };
89
90 static void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
91 {
92 spin_lock(&bdev->bd_size_lock);
93 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
94 spin_unlock(&bdev->bd_size_lock);
95 }
96
97 static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
98 {
99 struct parsed_partitions *state;
100 int nr;
101
102 state = kzalloc(sizeof(*state), GFP_KERNEL);
103 if (!state)
104 return NULL;
105
106 nr = disk_max_parts(hd);
107 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
108 if (!state->parts) {
109 kfree(state);
110 return NULL;
111 }
112
113 state->limit = nr;
114
115 return state;
116 }
117
118 static void free_partitions(struct parsed_partitions *state)
119 {
120 vfree(state->parts);
121 kfree(state);
122 }
123
124 static struct parsed_partitions *check_partition(struct gendisk *hd)
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->disk = hd;
140 snprintf(state->name, BDEVNAME_SIZE, "%s", hd->disk_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 put_disk(dev_to_bdev(dev)->bd_disk);
264 iput(dev_to_bdev(dev)->bd_inode);
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 static void delete_partition(struct block_device *part)
285 {
286 lockdep_assert_held(&part->bd_disk->open_mutex);
287
288 fsync_bdev(part);
289 __invalidate_device(part, true);
290
291 xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
292 kobject_put(part->bd_holder_dir);
293 device_del(&part->bd_device);
294
295 /*
296 * Remove the block device from the inode hash, so that it cannot be
297 * looked up any more even when openers still hold references.
298 */
299 remove_inode_hash(part->bd_inode);
300
301 put_device(&part->bd_device);
302 }
303
304 static ssize_t whole_disk_show(struct device *dev,
305 struct device_attribute *attr, char *buf)
306 {
307 return 0;
308 }
309 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
310
311 /*
312 * Must be called either with open_mutex held, before a disk can be opened or
313 * after all disk users are gone.
314 */
315 static struct block_device *add_partition(struct gendisk *disk, int partno,
316 sector_t start, sector_t len, int flags,
317 struct partition_meta_info *info)
318 {
319 dev_t devt = MKDEV(0, 0);
320 struct device *ddev = disk_to_dev(disk);
321 struct device *pdev;
322 struct block_device *bdev;
323 const char *dname;
324 int err;
325
326 lockdep_assert_held(&disk->open_mutex);
327
328 if (partno >= disk_max_parts(disk))
329 return ERR_PTR(-EINVAL);
330
331 /*
332 * Partitions are not supported on zoned block devices that are used as
333 * such.
334 */
335 switch (disk->queue->limits.zoned) {
336 case BLK_ZONED_HM:
337 pr_warn("%s: partitions not supported on host managed zoned block device\n",
338 disk->disk_name);
339 return ERR_PTR(-ENXIO);
340 case BLK_ZONED_HA:
341 pr_info("%s: disabling host aware zoned block device support due to partitions\n",
342 disk->disk_name);
343 blk_queue_set_zoned(disk, BLK_ZONED_NONE);
344 break;
345 case BLK_ZONED_NONE:
346 break;
347 }
348
349 if (xa_load(&disk->part_tbl, partno))
350 return ERR_PTR(-EBUSY);
351
352 /* ensure we always have a reference to the whole disk */
353 get_device(disk_to_dev(disk));
354
355 err = -ENOMEM;
356 bdev = bdev_alloc(disk, partno);
357 if (!bdev)
358 goto out_put_disk;
359
360 bdev->bd_start_sect = start;
361 bdev_set_nr_sectors(bdev, len);
362
363 pdev = &bdev->bd_device;
364 dname = dev_name(ddev);
365 if (isdigit(dname[strlen(dname) - 1]))
366 dev_set_name(pdev, "%sp%d", dname, partno);
367 else
368 dev_set_name(pdev, "%s%d", dname, partno);
369
370 device_initialize(pdev);
371 pdev->class = &block_class;
372 pdev->type = &part_type;
373 pdev->parent = ddev;
374
375 /* in consecutive minor range? */
376 if (bdev->bd_partno < disk->minors) {
377 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno);
378 } else {
379 err = blk_alloc_ext_minor();
380 if (err < 0)
381 goto out_put;
382 devt = MKDEV(BLOCK_EXT_MAJOR, err);
383 }
384 pdev->devt = devt;
385
386 if (info) {
387 err = -ENOMEM;
388 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
389 if (!bdev->bd_meta_info)
390 goto out_put;
391 }
392
393 /* delay uevent until 'holders' subdir is created */
394 dev_set_uevent_suppress(pdev, 1);
395 err = device_add(pdev);
396 if (err)
397 goto out_put;
398
399 err = -ENOMEM;
400 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
401 if (!bdev->bd_holder_dir)
402 goto out_del;
403
404 dev_set_uevent_suppress(pdev, 0);
405 if (flags & ADDPART_FLAG_WHOLEDISK) {
406 err = device_create_file(pdev, &dev_attr_whole_disk);
407 if (err)
408 goto out_del;
409 }
410
411 /* everything is up and running, commence */
412 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
413 if (err)
414 goto out_del;
415 bdev_add(bdev, devt);
416
417 /* suppress uevent if the disk suppresses it */
418 if (!dev_get_uevent_suppress(ddev))
419 kobject_uevent(&pdev->kobj, KOBJ_ADD);
420 return bdev;
421
422 out_del:
423 kobject_put(bdev->bd_holder_dir);
424 device_del(pdev);
425 out_put:
426 put_device(pdev);
427 return ERR_PTR(err);
428 out_put_disk:
429 put_disk(disk);
430 return ERR_PTR(err);
431 }
432
433 static bool partition_overlaps(struct gendisk *disk, sector_t start,
434 sector_t length, int skip_partno)
435 {
436 struct block_device *part;
437 bool overlap = false;
438 unsigned long idx;
439
440 rcu_read_lock();
441 xa_for_each_start(&disk->part_tbl, idx, part, 1) {
442 if (part->bd_partno != skip_partno &&
443 start < part->bd_start_sect + bdev_nr_sectors(part) &&
444 start + length > part->bd_start_sect) {
445 overlap = true;
446 break;
447 }
448 }
449 rcu_read_unlock();
450
451 return overlap;
452 }
453
454 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
455 sector_t length)
456 {
457 struct block_device *part;
458 int ret;
459
460 mutex_lock(&disk->open_mutex);
461 if (!disk_live(disk)) {
462 ret = -ENXIO;
463 goto out;
464 }
465
466 if (partition_overlaps(disk, start, length, -1)) {
467 ret = -EBUSY;
468 goto out;
469 }
470
471 part = add_partition(disk, partno, start, length,
472 ADDPART_FLAG_NONE, NULL);
473 ret = PTR_ERR_OR_ZERO(part);
474 out:
475 mutex_unlock(&disk->open_mutex);
476 return ret;
477 }
478
479 int bdev_del_partition(struct gendisk *disk, int partno)
480 {
481 struct block_device *part = NULL;
482 int ret = -ENXIO;
483
484 mutex_lock(&disk->open_mutex);
485 part = xa_load(&disk->part_tbl, partno);
486 if (!part)
487 goto out_unlock;
488
489 ret = -EBUSY;
490 if (part->bd_openers)
491 goto out_unlock;
492
493 delete_partition(part);
494 ret = 0;
495 out_unlock:
496 mutex_unlock(&disk->open_mutex);
497 return ret;
498 }
499
500 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
501 sector_t length)
502 {
503 struct block_device *part = NULL;
504 int ret = -ENXIO;
505
506 mutex_lock(&disk->open_mutex);
507 part = xa_load(&disk->part_tbl, partno);
508 if (!part)
509 goto out_unlock;
510
511 ret = -EINVAL;
512 if (start != part->bd_start_sect)
513 goto out_unlock;
514
515 ret = -EBUSY;
516 if (partition_overlaps(disk, start, length, partno))
517 goto out_unlock;
518
519 bdev_set_nr_sectors(part, length);
520
521 ret = 0;
522 out_unlock:
523 mutex_unlock(&disk->open_mutex);
524 return ret;
525 }
526
527 static bool disk_unlock_native_capacity(struct gendisk *disk)
528 {
529 const struct block_device_operations *bdops = disk->fops;
530
531 if (bdops->unlock_native_capacity &&
532 !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
533 printk(KERN_CONT "enabling native capacity\n");
534 bdops->unlock_native_capacity(disk);
535 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
536 return true;
537 } else {
538 printk(KERN_CONT "truncated\n");
539 return false;
540 }
541 }
542
543 void blk_drop_partitions(struct gendisk *disk)
544 {
545 struct block_device *part;
546 unsigned long idx;
547
548 lockdep_assert_held(&disk->open_mutex);
549
550 xa_for_each_start(&disk->part_tbl, idx, part, 1)
551 delete_partition(part);
552 }
553
554 static bool blk_add_partition(struct gendisk *disk,
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 static int blk_add_partitions(struct gendisk *disk)
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);
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, 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 int bdev_disk_changed(struct gendisk *disk, bool invalidate)
666 {
667 int ret = 0;
668
669 lockdep_assert_held(&disk->open_mutex);
670
671 if (!disk_live(disk))
672 return -ENXIO;
673
674 rescan:
675 if (disk->open_partitions)
676 return -EBUSY;
677 sync_blockdev(disk->part0);
678 invalidate_bdev(disk->part0);
679 blk_drop_partitions(disk);
680
681 clear_bit(GD_NEED_PART_SCAN, &disk->state);
682
683 /*
684 * Historically we only set the capacity to zero for devices that
685 * support partitions (independ of actually having partitions created).
686 * Doing that is rather inconsistent, but changing it broke legacy
687 * udisks polling for legacy ide-cdrom devices. Use the crude check
688 * below to get the sane behavior for most device while not breaking
689 * userspace for this particular setup.
690 */
691 if (invalidate) {
692 if (disk_part_scan_enabled(disk) ||
693 !(disk->flags & GENHD_FL_REMOVABLE))
694 set_capacity(disk, 0);
695 }
696
697 if (get_capacity(disk)) {
698 ret = blk_add_partitions(disk);
699 if (ret == -EAGAIN)
700 goto rescan;
701 } else if (invalidate) {
702 /*
703 * Tell userspace that the media / partition table may have
704 * changed.
705 */
706 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
707 }
708
709 return ret;
710 }
711 /*
712 * Only exported for loop and dasd for historic reasons. Don't use in new
713 * code!
714 */
715 EXPORT_SYMBOL_GPL(bdev_disk_changed);
716
717 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
718 {
719 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping;
720 struct page *page;
721
722 if (n >= get_capacity(state->disk)) {
723 state->access_beyond_eod = true;
724 return NULL;
725 }
726
727 page = read_mapping_page(mapping,
728 (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL);
729 if (IS_ERR(page))
730 goto out;
731 if (PageError(page))
732 goto out_put_page;
733
734 p->v = page;
735 return (unsigned char *)page_address(page) +
736 ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT);
737 out_put_page:
738 put_page(page);
739 out:
740 p->v = NULL;
741 return NULL;
742 }