projects/linux·blob master · refs

733 lines · 17.2 KiB

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/string.h>
11#include <linux/sysfs.h>
12#include <linux/ctype.h>
13#include <linux/vmalloc.h>
14#include <linux/raid/detect.h>
15#include "check.h"
16
17static int (*const 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_OF_PARTITION
49 of_partition, /* cmdline have priority to OF */
50#endif
51#ifdef CONFIG_EFI_PARTITION
52 efi_partition, /* this must come before msdos */
53#endif
54#ifdef CONFIG_SGI_PARTITION
55 sgi_partition,
56#endif
57#ifdef CONFIG_LDM_PARTITION
58 ldm_partition, /* this must come before msdos */
59#endif
60#ifdef CONFIG_MSDOS_PARTITION
61 msdos_partition,
62#endif
63#ifdef CONFIG_OSF_PARTITION
64 osf_partition,
65#endif
66#ifdef CONFIG_SUN_PARTITION
67 sun_partition,
68#endif
69#ifdef CONFIG_AMIGA_PARTITION
70 amiga_partition,
71#endif
72#ifdef CONFIG_ATARI_PARTITION
73 atari_partition,
74#endif
75#ifdef CONFIG_MAC_PARTITION
76 mac_partition,
77#endif
78#ifdef CONFIG_ULTRIX_PARTITION
79 ultrix_partition,
80#endif
81#ifdef CONFIG_IBM_PARTITION
82 ibm_partition,
83#endif
84#ifdef CONFIG_KARMA_PARTITION
85 karma_partition,
86#endif
87#ifdef CONFIG_SYSV68_PARTITION
88 sysv68_partition,
89#endif
90 NULL
91};
92
93static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
94{
95 struct parsed_partitions *state;
96 int nr = DISK_MAX_PARTS;
97
98 state = kzalloc_obj(*state);
99 if (!state)
100 return NULL;
101
102 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
103 if (!state->parts) {
104 kfree(state);
105 return NULL;
106 }
107
108 state->limit = nr;
109
110 return state;
111}
112
113static void free_partitions(struct parsed_partitions *state)
114{
115 vfree(state->parts);
116 kfree(state);
117}
118
119static struct parsed_partitions *check_partition(struct gendisk *hd)
120{
121 struct parsed_partitions *state;
122 int i, res, err;
123
124 state = allocate_partitions(hd);
125 if (!state)
126 return NULL;
127 state->pp_buf.buffer = kmalloc(PAGE_SIZE, GFP_KERNEL);
128 if (!state->pp_buf.buffer) {
129 free_partitions(state);
130 return NULL;
131 }
132 seq_buf_init(&state->pp_buf, state->pp_buf.buffer, PAGE_SIZE);
133
134 state->disk = hd;
135 strscpy(state->name, hd->disk_name);
136 seq_buf_printf(&state->pp_buf, " %s:", state->name);
137 if (isdigit(state->name[strlen(state->name)-1]))
138 sprintf(state->name, "p");
139
140 i = res = err = 0;
141 while (!res && check_part[i]) {
142 memset(state->parts, 0, state->limit * sizeof(state->parts[0]));
143 res = check_part[i++](state);
144 if (res < 0) {
145 /*
146 * We have hit an I/O error which we don't report now.
147 * But record it, and let the others do their job.
148 */
149 err = res;
150 res = 0;
151 }
152
153 }
154 if (res > 0) {
155 printk(KERN_INFO "%s", seq_buf_str(&state->pp_buf));
156
157 kfree(state->pp_buf.buffer);
158 return state;
159 }
160 if (state->access_beyond_eod)
161 err = -ENOSPC;
162 /*
163 * The partition is unrecognized. So report I/O errors if there were any
164 */
165 if (err)
166 res = err;
167 if (res) {
168 seq_buf_puts(&state->pp_buf,
169 " unable to read partition table\n");
170 printk(KERN_INFO "%s", seq_buf_str(&state->pp_buf));
171 }
172
173 kfree(state->pp_buf.buffer);
174 free_partitions(state);
175 return ERR_PTR(res);
176}
177
178static ssize_t part_partition_show(struct device *dev,
179 struct device_attribute *attr, char *buf)
180{
181 return sysfs_emit(buf, "%d\n", bdev_partno(dev_to_bdev(dev)));
182}
183
184static ssize_t part_start_show(struct device *dev,
185 struct device_attribute *attr, char *buf)
186{
187 return sysfs_emit(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect);
188}
189
190static ssize_t part_ro_show(struct device *dev,
191 struct device_attribute *attr, char *buf)
192{
193 return sysfs_emit(buf, "%d\n", bdev_read_only(dev_to_bdev(dev)));
194}
195
196static ssize_t part_alignment_offset_show(struct device *dev,
197 struct device_attribute *attr, char *buf)
198{
199 return sysfs_emit(buf, "%u\n", bdev_alignment_offset(dev_to_bdev(dev)));
200}
201
202static ssize_t part_discard_alignment_show(struct device *dev,
203 struct device_attribute *attr, char *buf)
204{
205 return sysfs_emit(buf, "%u\n", bdev_discard_alignment(dev_to_bdev(dev)));
206}
207
208static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
209static DEVICE_ATTR(start, 0444, part_start_show, NULL);
210static DEVICE_ATTR(size, 0444, part_size_show, NULL);
211static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
212static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
213static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
214static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
215static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
216#ifdef CONFIG_FAIL_MAKE_REQUEST
217static struct device_attribute dev_attr_fail =
218 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
219#endif
220
221static struct attribute *part_attrs[] = {
222 &dev_attr_partition.attr,
223 &dev_attr_start.attr,
224 &dev_attr_size.attr,
225 &dev_attr_ro.attr,
226 &dev_attr_alignment_offset.attr,
227 &dev_attr_discard_alignment.attr,
228 &dev_attr_stat.attr,
229 &dev_attr_inflight.attr,
230#ifdef CONFIG_FAIL_MAKE_REQUEST
231 &dev_attr_fail.attr,
232#endif
233 NULL
234};
235
236static const struct attribute_group part_attr_group = {
237 .attrs = part_attrs,
238};
239
240static const struct attribute_group *part_attr_groups[] = {
241 &part_attr_group,
242#ifdef CONFIG_BLK_DEV_IO_TRACE
243 &blk_trace_attr_group,
244#endif
245 NULL
246};
247
248static void part_release(struct device *dev)
249{
250 put_disk(dev_to_bdev(dev)->bd_disk);
251 bdev_drop(dev_to_bdev(dev));
252}
253
254static int part_uevent(const struct device *dev, struct kobj_uevent_env *env)
255{
256 const struct block_device *part = dev_to_bdev(dev);
257
258 add_uevent_var(env, "PARTN=%u", bdev_partno(part));
259 if (part->bd_meta_info && part->bd_meta_info->volname[0])
260 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname);
261 if (part->bd_meta_info && part->bd_meta_info->uuid[0])
262 add_uevent_var(env, "PARTUUID=%s", part->bd_meta_info->uuid);
263 return 0;
264}
265
266const struct device_type part_type = {
267 .name = "partition",
268 .groups = part_attr_groups,
269 .release = part_release,
270 .uevent = part_uevent,
271};
272
273void drop_partition(struct block_device *part)
274{
275 lockdep_assert_held(&part->bd_disk->open_mutex);
276
277 xa_erase(&part->bd_disk->part_tbl, bdev_partno(part));
278 kobject_put(part->bd_holder_dir);
279
280 device_del(&part->bd_device);
281 put_device(&part->bd_device);
282}
283
284static ssize_t whole_disk_show(struct device *dev,
285 struct device_attribute *attr, char *buf)
286{
287 return 0;
288}
289static const DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
290
291/*
292 * Must be called either with open_mutex held, before a disk can be opened or
293 * after all disk users are gone.
294 */
295static struct block_device *add_partition(struct gendisk *disk, int partno,
296 sector_t start, sector_t len, int flags,
297 struct partition_meta_info *info)
298{
299 dev_t devt = MKDEV(0, 0);
300 struct device *ddev = disk_to_dev(disk);
301 struct device *pdev;
302 struct block_device *bdev;
303 const char *dname;
304 int err;
305
306 lockdep_assert_held(&disk->open_mutex);
307
308 if (partno >= DISK_MAX_PARTS)
309 return ERR_PTR(-EINVAL);
310
311 /*
312 * Partitions are not supported on zoned block devices that are used as
313 * such.
314 */
315 if (bdev_is_zoned(disk->part0)) {
316 pr_warn("%s: partitions not supported on host managed zoned block device\n",
317 disk->disk_name);
318 return ERR_PTR(-ENXIO);
319 }
320
321 if (xa_load(&disk->part_tbl, partno))
322 return ERR_PTR(-EBUSY);
323
324 /* ensure we always have a reference to the whole disk */
325 get_device(disk_to_dev(disk));
326
327 err = -ENOMEM;
328 bdev = bdev_alloc(disk, partno);
329 if (!bdev)
330 goto out_put_disk;
331
332 bdev->bd_start_sect = start;
333 bdev_set_nr_sectors(bdev, len);
334
335 pdev = &bdev->bd_device;
336 dname = dev_name(ddev);
337 if (isdigit(dname[strlen(dname) - 1]))
338 dev_set_name(pdev, "%sp%d", dname, partno);
339 else
340 dev_set_name(pdev, "%s%d", dname, partno);
341
342 device_initialize(pdev);
343 pdev->class = &block_class;
344 pdev->type = &part_type;
345 pdev->parent = ddev;
346
347 /* in consecutive minor range? */
348 if (bdev_partno(bdev) < disk->minors) {
349 devt = MKDEV(disk->major, disk->first_minor + bdev_partno(bdev));
350 } else {
351 err = blk_alloc_ext_minor();
352 if (err < 0)
353 goto out_put;
354 devt = MKDEV(BLOCK_EXT_MAJOR, err);
355 }
356 pdev->devt = devt;
357
358 if (info) {
359 err = -ENOMEM;
360 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
361 if (!bdev->bd_meta_info)
362 goto out_put;
363 }
364
365 /* delay uevent until 'holders' subdir is created */
366 dev_set_uevent_suppress(pdev, 1);
367 err = device_add(pdev);
368 if (err)
369 goto out_put;
370
371 err = -ENOMEM;
372 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
373 if (!bdev->bd_holder_dir)
374 goto out_del;
375
376 dev_set_uevent_suppress(pdev, 0);
377 if (flags & ADDPART_FLAG_WHOLEDISK) {
378 err = device_create_file(pdev, &dev_attr_whole_disk);
379 if (err)
380 goto out_del;
381 }
382
383 if (flags & ADDPART_FLAG_READONLY)
384 bdev_set_flag(bdev, BD_READ_ONLY);
385
386 /* everything is up and running, commence */
387 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL);
388 if (err)
389 goto out_del;
390 bdev_add(bdev, devt);
391
392 /* suppress uevent if the disk suppresses it */
393 if (!dev_get_uevent_suppress(ddev))
394 kobject_uevent(&pdev->kobj, KOBJ_ADD);
395 return bdev;
396
397out_del:
398 kobject_put(bdev->bd_holder_dir);
399 device_del(pdev);
400out_put:
401 put_device(pdev);
402 return ERR_PTR(err);
403out_put_disk:
404 put_disk(disk);
405 return ERR_PTR(err);
406}
407
408static bool partition_overlaps(struct gendisk *disk, sector_t start,
409 sector_t length, int skip_partno)
410{
411 struct block_device *part;
412 bool overlap = false;
413 unsigned long idx;
414
415 rcu_read_lock();
416 xa_for_each_start(&disk->part_tbl, idx, part, 1) {
417 if (bdev_partno(part) != skip_partno &&
418 start < part->bd_start_sect + bdev_nr_sectors(part) &&
419 start + length > part->bd_start_sect) {
420 overlap = true;
421 break;
422 }
423 }
424 rcu_read_unlock();
425
426 return overlap;
427}
428
429int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
430 sector_t length)
431{
432 struct block_device *part;
433 int ret;
434
435 mutex_lock(&disk->open_mutex);
436 if (!disk_live(disk)) {
437 ret = -ENXIO;
438 goto out;
439 }
440
441 if (disk->flags & GENHD_FL_NO_PART) {
442 ret = -EINVAL;
443 goto out;
444 }
445
446 if (partition_overlaps(disk, start, length, -1)) {
447 ret = -EBUSY;
448 goto out;
449 }
450
451 part = add_partition(disk, partno, start, length,
452 ADDPART_FLAG_NONE, NULL);
453 ret = PTR_ERR_OR_ZERO(part);
454out:
455 mutex_unlock(&disk->open_mutex);
456 return ret;
457}
458
459int bdev_del_partition(struct gendisk *disk, int partno)
460{
461 struct block_device *part = NULL;
462 int ret = -ENXIO;
463
464 mutex_lock(&disk->open_mutex);
465 part = xa_load(&disk->part_tbl, partno);
466 if (!part)
467 goto out_unlock;
468
469 ret = -EBUSY;
470 if (atomic_read(&part->bd_openers))
471 goto out_unlock;
472
473 /*
474 * We verified that @part->bd_openers is zero above and so
475 * @part->bd_holder{_ops} can't be set. And since we hold
476 * @disk->open_mutex the device can't be claimed by anyone.
477 *
478 * So no need to call @part->bd_holder_ops->mark_dead() here.
479 * Just delete the partition and invalidate it.
480 */
481
482 bdev_unhash(part);
483 invalidate_bdev(part);
484 drop_partition(part);
485 ret = 0;
486out_unlock:
487 mutex_unlock(&disk->open_mutex);
488 return ret;
489}
490
491int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
492 sector_t length)
493{
494 struct block_device *part = NULL;
495 int ret = -ENXIO;
496
497 mutex_lock(&disk->open_mutex);
498 part = xa_load(&disk->part_tbl, partno);
499 if (!part)
500 goto out_unlock;
501
502 ret = -EINVAL;
503 if (start != part->bd_start_sect)
504 goto out_unlock;
505
506 ret = -EBUSY;
507 if (partition_overlaps(disk, start, length, partno))
508 goto out_unlock;
509
510 bdev_set_nr_sectors(part, length);
511
512 ret = 0;
513out_unlock:
514 mutex_unlock(&disk->open_mutex);
515 return ret;
516}
517
518static bool disk_unlock_native_capacity(struct gendisk *disk)
519{
520 if (!disk->fops->unlock_native_capacity ||
521 test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) {
522 printk(KERN_CONT "truncated\n");
523 return false;
524 }
525
526 printk(KERN_CONT "enabling native capacity\n");
527 disk->fops->unlock_native_capacity(disk);
528 return true;
529}
530
531static bool blk_add_partition(struct gendisk *disk,
532 struct parsed_partitions *state, int p)
533{
534 sector_t size = state->parts[p].size;
535 sector_t from = state->parts[p].from;
536 struct block_device *part;
537
538 if (!size)
539 return true;
540
541 if (from >= get_capacity(disk)) {
542 printk(KERN_WARNING
543 "%s: p%d start %llu is beyond EOD, ",
544 disk->disk_name, p, (unsigned long long) from);
545 if (disk_unlock_native_capacity(disk))
546 return false;
547 return true;
548 }
549
550 if (from + size > get_capacity(disk)) {
551 printk(KERN_WARNING
552 "%s: p%d size %llu extends beyond EOD, ",
553 disk->disk_name, p, (unsigned long long) size);
554
555 if (disk_unlock_native_capacity(disk))
556 return false;
557
558 /*
559 * We can not ignore partitions of broken tables created by for
560 * example camera firmware, but we limit them to the end of the
561 * disk to avoid creating invalid block devices.
562 */
563 size = get_capacity(disk) - from;
564 }
565
566 part = add_partition(disk, p, from, size, state->parts[p].flags,
567 &state->parts[p].info);
568 if (IS_ERR(part)) {
569 if (PTR_ERR(part) != -ENXIO) {
570 printk(KERN_ERR " %s: p%d could not be added: %pe\n",
571 disk->disk_name, p, part);
572 }
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
583static int blk_add_partitions(struct gendisk *disk)
584{
585 struct parsed_partitions *state;
586 int ret = -EAGAIN, p;
587
588 if (!disk_has_partscan(disk))
589 return 0;
590
591 state = check_partition(disk);
592 if (!state)
593 return 0;
594 if (IS_ERR(state)) {
595 /*
596 * I/O error reading the partition table. If we tried to read
597 * beyond EOD, retry after unlocking the native capacity.
598 */
599 if (PTR_ERR(state) == -ENOSPC) {
600 printk(KERN_WARNING "%s: partition table beyond EOD, ",
601 disk->disk_name);
602 if (disk_unlock_native_capacity(disk))
603 return -EAGAIN;
604 }
605 return -EIO;
606 }
607
608 /*
609 * Partitions are not supported on host managed zoned block devices.
610 */
611 if (bdev_is_zoned(disk->part0)) {
612 pr_warn("%s: ignoring partition table on host managed zoned block device\n",
613 disk->disk_name);
614 ret = 0;
615 goto out_free_state;
616 }
617
618 /*
619 * If we read beyond EOD, try unlocking native capacity even if the
620 * partition table was successfully read as we could be missing some
621 * partitions.
622 */
623 if (state->access_beyond_eod) {
624 printk(KERN_WARNING
625 "%s: partition table partially beyond EOD, ",
626 disk->disk_name);
627 if (disk_unlock_native_capacity(disk))
628 goto out_free_state;
629 }
630
631 /* tell userspace that the media / partition table may have changed */
632 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
633
634 for (p = 1; p < state->limit; p++)
635 if (!blk_add_partition(disk, state, p))
636 goto out_free_state;
637
638 ret = 0;
639out_free_state:
640 free_partitions(state);
641 return ret;
642}
643
644int bdev_disk_changed(struct gendisk *disk, bool invalidate)
645{
646 struct block_device *part;
647 unsigned long idx;
648 int ret = 0;
649
650 lockdep_assert_held(&disk->open_mutex);
651
652 if (!disk_live(disk))
653 return -ENXIO;
654
655rescan:
656 if (disk->open_partitions)
657 return -EBUSY;
658 sync_blockdev(disk->part0);
659 invalidate_bdev(disk->part0);
660
661 xa_for_each_start(&disk->part_tbl, idx, part, 1) {
662 /*
663 * Remove the block device from the inode hash, so that
664 * it cannot be looked up any more even when openers
665 * still hold references.
666 */
667 bdev_unhash(part);
668
669 /*
670 * If @disk->open_partitions isn't elevated but there's
671 * still an active holder of that block device things
672 * are broken.
673 */
674 WARN_ON_ONCE(atomic_read(&part->bd_openers));
675 invalidate_bdev(part);
676 drop_partition(part);
677 }
678 clear_bit(GD_NEED_PART_SCAN, &disk->state);
679
680 /*
681 * Historically we only set the capacity to zero for devices that
682 * support partitions (independ of actually having partitions created).
683 * Doing that is rather inconsistent, but changing it broke legacy
684 * udisks polling for legacy ide-cdrom devices. Use the crude check
685 * below to get the sane behavior for most device while not breaking
686 * userspace for this particular setup.
687 */
688 if (invalidate) {
689 if (!(disk->flags & GENHD_FL_NO_PART) ||
690 !(disk->flags & GENHD_FL_REMOVABLE))
691 set_capacity(disk, 0);
692 }
693
694 if (get_capacity(disk)) {
695 ret = blk_add_partitions(disk);
696 if (ret == -EAGAIN)
697 goto rescan;
698 } else if (invalidate) {
699 /*
700 * Tell userspace that the media / partition table may have
701 * changed.
702 */
703 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
704 }
705
706 return ret;
707}
708/*
709 * Only exported for loop and dasd for historic reasons. Don't use in new
710 * code!
711 */
712EXPORT_SYMBOL_GPL(bdev_disk_changed);
713
714void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
715{
716 struct address_space *mapping = state->disk->part0->bd_mapping;
717 struct folio *folio;
718
719 if (n >= get_capacity(state->disk)) {
720 state->access_beyond_eod = true;
721 goto out;
722 }
723
724 folio = read_mapping_folio(mapping, n >> PAGE_SECTORS_SHIFT, NULL);
725 if (IS_ERR(folio))
726 goto out;
727
728 p->v = folio;
729 return folio_address(folio) + offset_in_folio(folio, n * SECTOR_SIZE);
730out:
731 p->v = NULL;
732 return NULL;
733}