md/md-bitmap: Synchronize bitmap_get_stats() with bitmap lifetime
authorYu Kuai <yukuai3@huawei.com>
Fri, 24 Jan 2025 09:20:55 +0000 (17:20 +0800)
committerSong Liu <song@kernel.org>
Fri, 24 Jan 2025 18:03:32 +0000 (10:03 -0800)
After commit ec6bb299c7c3 ("md/md-bitmap: add 'sync_size' into struct
md_bitmap_stats"), following panic is reported:

Oops: general protection fault, probably for non-canonical address
RIP: 0010:bitmap_get_stats+0x2b/0xa0
Call Trace:
 <TASK>
 md_seq_show+0x2d2/0x5b0
 seq_read_iter+0x2b9/0x470
 seq_read+0x12f/0x180
 proc_reg_read+0x57/0xb0
 vfs_read+0xf6/0x380
 ksys_read+0x6c/0xf0
 do_syscall_64+0x82/0x170
 entry_SYSCALL_64_after_hwframe+0x76/0x7e

Root cause is that bitmap_get_stats() can be called at anytime if mddev
is still there, even if bitmap is destroyed, or not fully initialized.
Deferenceing bitmap in this case can crash the kernel. Meanwhile, the
above commit start to deferencing bitmap->storage, make the problem
easier to trigger.

Fix the problem by protecting bitmap_get_stats() with bitmap_info.mutex.

Cc: stable@vger.kernel.org # v6.12+
Fixes: 32a7627cf3a3 ("[PATCH] md: optimised resync using Bitmap based intent logging")
Reported-and-tested-by: Harshit Mogalapalli <harshit.m.mogalapalli@oracle.com>
Closes: https://lore.kernel.org/linux-raid/ca3a91a2-50ae-4f68-b317-abd9889f3907@oracle.com/T/#m6e5086c95201135e4941fe38f9efa76daf9666c5
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20250124092055.4050195-1-yukuai1@huaweicloud.com
Signed-off-by: Song Liu <song@kernel.org>
drivers/md/md-bitmap.c
drivers/md/md.c

index ec4ecd96e6b14c478c4d0a596fe6e0a4a70f52ad..23c09d22fcdbc19e76f2a0a229ec890779115545 100644 (file)
@@ -2355,7 +2355,10 @@ static int bitmap_get_stats(void *data, struct md_bitmap_stats *stats)
 
        if (!bitmap)
                return -ENOENT;
-
+       if (bitmap->mddev->bitmap_info.external)
+               return -ENOENT;
+       if (!bitmap->storage.sb_page) /* no superblock */
+               return -EINVAL;
        sb = kmap_local_page(bitmap->storage.sb_page);
        stats->sync_size = le64_to_cpu(sb->sync_size);
        kunmap_local(sb);
index 866015b681af8e95a160f1a3bf6058b28bc99906..465ca2af1e6efbfb931857b2dc9464be5626d857 100644 (file)
@@ -8376,6 +8376,10 @@ static int md_seq_show(struct seq_file *seq, void *v)
                return 0;
 
        spin_unlock(&all_mddevs_lock);
+
+       /* prevent bitmap to be freed after checking */
+       mutex_lock(&mddev->bitmap_info.mutex);
+
        spin_lock(&mddev->lock);
        if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
                seq_printf(seq, "%s : ", mdname(mddev));
@@ -8451,6 +8455,7 @@ static int md_seq_show(struct seq_file *seq, void *v)
                seq_printf(seq, "\n");
        }
        spin_unlock(&mddev->lock);
+       mutex_unlock(&mddev->bitmap_info.mutex);
        spin_lock(&all_mddevs_lock);
 
        if (mddev == list_last_entry(&all_mddevs, struct mddev, all_mddevs))