Pull device mapper fixes from Mike Snitzer: - Fix DM core retrieve_deps() UAF race due to missing locking of a DM table's list of devices that is managed using dm_{get,put}_device. - Revert DM core's half-baked RCU optimization if IO submitter has set REQ_NOWAIT. Can be revisited, and properly justified, after comprehensively auditing all of DM to also pass GFP_NOWAIT for any allocations if REQ_NOWAIT used. * tag 'for-6.6/dm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm: dm: don't attempt to queue IO under RCU protection dm: fix a race condition in retrieve_deps
Unified · Split
| @@ -214,6 +214,7 @@ struct dm_table { · 213 unmodified lines | |||
| 214 | 214 | ||
| 215 | /* a list of devices used by this table */ | 215 | /* a list of devices used by this table */ |
| 216 | struct list_head devices; | 216 | struct list_head devices; |
| 217 | struct rw_semaphore devices_lock; | ||
| 217 | 218 | ||
| 218 | /* events get handed up using this callback */ | 219 | /* events get handed up using this callback */ |
| 219 | void (*event_fn)(void *data); | 220 | void (*event_fn)(void *data); |
| @@ -1630,6 +1630,8 @@ static void retrieve_deps(struct dm_table *table, · 1629 unmodified lines | |||
| 1630 | struct dm_dev_internal *dd; | 1630 | struct dm_dev_internal *dd; |
| 1631 | struct dm_target_deps *deps; | 1631 | struct dm_target_deps *deps; |
| 1632 | 1632 | ||
| 1633 | down_read(&table->devices_lock); | ||
| 1634 | |||
| 1633 | deps = get_result_buffer(param, param_size, &len); | 1635 | deps = get_result_buffer(param, param_size, &len); |
| 1634 | 1636 | ||
| 1635 | /* | 1637 | /* |
| @@ -1644,7 +1646,7 @@ static void retrieve_deps(struct dm_table *table, · 8 unmodified lines | |||
| 1644 | needed = struct_size(deps, dev, count); | 1646 | needed = struct_size(deps, dev, count); |
| 1645 | if (len < needed) { | 1647 | if (len < needed) { |
| 1646 | param->flags |= DM_BUFFER_FULL_FLAG; | 1648 | param->flags |= DM_BUFFER_FULL_FLAG; |
| 1647 | return; | 1649 | goto out; |
| 1648 | } | 1650 | } |
| 1649 | 1651 | ||
| 1650 | /* | 1652 | /* |
| @@ -1656,6 +1658,9 @@ static void retrieve_deps(struct dm_table *table, · 5 unmodified lines | |||
| 1656 | deps->dev[count++] = huge_encode_dev(dd->dm_dev->bdev->bd_dev); | 1658 | deps->dev[count++] = huge_encode_dev(dd->dm_dev->bdev->bd_dev); |
| 1657 | 1659 | ||
| 1658 | param->data_size = param->data_start + needed; | 1660 | param->data_size = param->data_start + needed; |
| 1661 | |||
| 1662 | out: | ||
| 1663 | up_read(&table->devices_lock); | ||
| 1659 | } | 1664 | } |
| 1660 | 1665 | ||
| 1661 | static int table_deps(struct file *filp, struct dm_ioctl *param, size_t param_size) | 1666 | static int table_deps(struct file *filp, struct dm_ioctl *param, size_t param_size) |
| @@ -135,6 +135,7 @@ int dm_table_create(struct dm_table **result, blk_mode_t mode, · 134 unmodified lines | |||
| 135 | return -ENOMEM; | 135 | return -ENOMEM; |
| 136 | 136 | ||
| 137 | INIT_LIST_HEAD(&t->devices); | 137 | INIT_LIST_HEAD(&t->devices); |
| 138 | init_rwsem(&t->devices_lock); | ||
| 138 | 139 | ||
| 139 | if (!num_targets) | 140 | if (!num_targets) |
| 140 | num_targets = KEYS_PER_NODE; | 141 | num_targets = KEYS_PER_NODE; |
| @@ -359,16 +360,20 @@ int __ref dm_get_device(struct dm_target *ti, const char *path, blk_mode_t mode, · 218 unmodified lines | |||
| 359 | if (dev == disk_devt(t->md->disk)) | 360 | if (dev == disk_devt(t->md->disk)) |
| 360 | return -EINVAL; | 361 | return -EINVAL; |
| 361 | 362 | ||
| 363 | down_write(&t->devices_lock); | ||
| 364 | |||
| 362 | dd = find_device(&t->devices, dev); | 365 | dd = find_device(&t->devices, dev); |
| 363 | if (!dd) { | 366 | if (!dd) { |
| 364 | dd = kmalloc(sizeof(*dd), GFP_KERNEL); | 367 | dd = kmalloc(sizeof(*dd), GFP_KERNEL); |
| 365 | if (!dd) | 368 | if (!dd) { |
| 366 | return -ENOMEM; | 369 | r = -ENOMEM; |
| 370 | goto unlock_ret_r; | ||
| 371 | } | ||
| 367 | 372 | ||
| 368 | r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev); | 373 | r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev); |
| 369 | if (r) { | 374 | if (r) { |
| 370 | kfree(dd); | 375 | kfree(dd); |
| 371 | return r; | 376 | goto unlock_ret_r; |
| 372 | } | 377 | } |
| 373 | 378 | ||
| 374 | refcount_set(&dd->count, 1); | 379 | refcount_set(&dd->count, 1); |
| @@ -378,12 +383,17 @@ int __ref dm_get_device(struct dm_target *ti, const char *path, blk_mode_t mode, · 3 unmodified lines | |||
| 378 | } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) { | 383 | } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) { |
| 379 | r = upgrade_mode(dd, mode, t->md); | 384 | r = upgrade_mode(dd, mode, t->md); |
| 380 | if (r) | 385 | if (r) |
| 381 | return r; | 386 | goto unlock_ret_r; |
| 382 | } | 387 | } |
| 383 | refcount_inc(&dd->count); | 388 | refcount_inc(&dd->count); |
| 384 | out: | 389 | out: |
| 390 | up_write(&t->devices_lock); | ||
| 385 | *result = dd->dm_dev; | 391 | *result = dd->dm_dev; |
| 386 | return 0; | 392 | return 0; |
| 393 | |||
| 394 | unlock_ret_r: | ||
| 395 | up_write(&t->devices_lock); | ||
| 396 | return r; | ||
| 387 | } | 397 | } |
| 388 | EXPORT_SYMBOL(dm_get_device); | 398 | EXPORT_SYMBOL(dm_get_device); |
| 389 | 399 | ||
| @@ -419,9 +429,12 @@ static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev, · 29 unmodified lines | |||
| 419 | void dm_put_device(struct dm_target *ti, struct dm_dev *d) | 429 | void dm_put_device(struct dm_target *ti, struct dm_dev *d) |
| 420 | { | 430 | { |
| 421 | int found = 0; | 431 | int found = 0; |
| 422 | struct list_head *devices = &ti->table->devices; | 432 | struct dm_table *t = ti->table; |
| 433 | struct list_head *devices = &t->devices; | ||
| 423 | struct dm_dev_internal *dd; | 434 | struct dm_dev_internal *dd; |
| 424 | 435 | ||
| 436 | down_write(&t->devices_lock); | ||
| 437 | |||
| 425 | list_for_each_entry(dd, devices, list) { | 438 | list_for_each_entry(dd, devices, list) { |
| 426 | if (dd->dm_dev == d) { | 439 | if (dd->dm_dev == d) { |
| 427 | found = 1; | 440 | found = 1; |
| @@ -430,14 +443,17 @@ void dm_put_device(struct dm_target *ti, struct dm_dev *d) · 2 unmodified lines | |||
| 430 | } | 443 | } |
| 431 | if (!found) { | 444 | if (!found) { |
| 432 | DMERR("%s: device %s not in table devices list", | 445 | DMERR("%s: device %s not in table devices list", |
| 433 | dm_device_name(ti->table->md), d->name); | 446 | dm_device_name(t->md), d->name); |
| 434 | return; | 447 | goto unlock_ret; |
| 435 | } | 448 | } |
| 436 | if (refcount_dec_and_test(&dd->count)) { | 449 | if (refcount_dec_and_test(&dd->count)) { |
| 437 | dm_put_table_device(ti->table->md, d); | 450 | dm_put_table_device(t->md, d); |
| 438 | list_del(&dd->list); | 451 | list_del(&dd->list); |
| 439 | kfree(dd); | 452 | kfree(dd); |
| 440 | } | 453 | } |
| 454 | |||
| 455 | unlock_ret: | ||
| 456 | up_write(&t->devices_lock); | ||
| 441 | } | 457 | } |
| 442 | EXPORT_SYMBOL(dm_put_device); | 458 | EXPORT_SYMBOL(dm_put_device); |
| 443 | 459 | ||
| @@ -715,24 +715,6 @@ static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) · 714 unmodified lines | |||
| 715 | rcu_read_unlock(); | 715 | rcu_read_unlock(); |
| 716 | } | 716 | } |
| 717 | 717 | ||
| 718 | static inline struct dm_table *dm_get_live_table_bio(struct mapped_device *md, | ||
| 719 | int *srcu_idx, blk_opf_t bio_opf) | ||
| 720 | { | ||
| 721 | if (bio_opf & REQ_NOWAIT) | ||
| 722 | return dm_get_live_table_fast(md); | ||
| 723 | else | ||
| 724 | return dm_get_live_table(md, srcu_idx); | ||
| 725 | } | ||
| 726 | |||
| 727 | static inline void dm_put_live_table_bio(struct mapped_device *md, int srcu_idx, | ||
| 728 | blk_opf_t bio_opf) | ||
| 729 | { | ||
| 730 | if (bio_opf & REQ_NOWAIT) | ||
| 731 | dm_put_live_table_fast(md); | ||
| 732 | else | ||
| 733 | dm_put_live_table(md, srcu_idx); | ||
| 734 | } | ||
| 735 | |||
| 736 | static char *_dm_claim_ptr = "I belong to device-mapper"; | 718 | static char *_dm_claim_ptr = "I belong to device-mapper"; |
| 737 | 719 | ||
| 738 | /* | 720 | /* |
| @@ -1833,9 +1815,8 @@ static void dm_submit_bio(struct bio *bio) · 1094 unmodified lines | |||
| 1833 | struct mapped_device *md = bio->bi_bdev->bd_disk->private_data; | 1815 | struct mapped_device *md = bio->bi_bdev->bd_disk->private_data; |
| 1834 | int srcu_idx; | 1816 | int srcu_idx; |
| 1835 | struct dm_table *map; | 1817 | struct dm_table *map; |
| 1836 | blk_opf_t bio_opf = bio->bi_opf; | ||
| 1837 | 1818 | ||
| 1838 | map = dm_get_live_table_bio(md, &srcu_idx, bio_opf); | 1819 | map = dm_get_live_table(md, &srcu_idx); |
| 1839 | 1820 | ||
| 1840 | /* If suspended, or map not yet available, queue this IO for later */ | 1821 | /* If suspended, or map not yet available, queue this IO for later */ |
| 1841 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) || | 1822 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) || |
| @@ -1851,7 +1832,7 @@ static void dm_submit_bio(struct bio *bio) · 9 unmodified lines | |||
| 1851 | 1832 | ||
| 1852 | dm_split_and_process_bio(md, map, bio); | 1833 | dm_split_and_process_bio(md, map, bio); |
| 1853 | out: | 1834 | out: |
| 1854 | dm_put_live_table_bio(md, srcu_idx, bio_opf); | 1835 | dm_put_live_table(md, srcu_idx); |
| 1855 | } | 1836 | } |
| 1856 | 1837 | ||
| 1857 | static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob, | 1838 | static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob, |