]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/commitdiff
btrfs: Wait for in-flight bios before freeing target device for raid56
authorQu Wenruo <quwenruo@cn.fujitsu.com>
Wed, 29 Mar 2017 01:33:21 +0000 (09:33 +0800)
committerDavid Sterba <dsterba@suse.com>
Tue, 18 Apr 2017 12:07:26 +0000 (14:07 +0200)
When raid56 dev-replace is cancelled by running scrub, we will free
target device without waiting for in-flight bios, causing the following
NULL pointer deference or general protection failure.

 BUG: unable to handle kernel NULL pointer dereference at 00000000000005e0
 IP: generic_make_request_checks+0x4d/0x610
 CPU: 1 PID: 11676 Comm: kworker/u4:14 Tainted: G  O    4.11.0-rc2 #72
 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.10.2-20170228_101828-anatol 04/01/2014
 Workqueue: btrfs-endio-raid56 btrfs_endio_raid56_helper [btrfs]
 task: ffff88002875b4c0 task.stack: ffffc90001334000
 RIP: 0010:generic_make_request_checks+0x4d/0x610
 Call Trace:
  ? generic_make_request+0xc7/0x360
  generic_make_request+0x24/0x360
  ? generic_make_request+0xc7/0x360
  submit_bio+0x64/0x120
  ? page_in_rbio+0x4d/0x80 [btrfs]
  ? rbio_orig_end_io+0x80/0x80 [btrfs]
  finish_rmw+0x3f4/0x540 [btrfs]
  validate_rbio_for_rmw+0x36/0x40 [btrfs]
  raid_rmw_end_io+0x7a/0x90 [btrfs]
  bio_endio+0x56/0x60
  end_workqueue_fn+0x3c/0x40 [btrfs]
  btrfs_scrubparity_helper+0xef/0x620 [btrfs]
  btrfs_endio_raid56_helper+0xe/0x10 [btrfs]
  process_one_work+0x2af/0x720
  ? process_one_work+0x22b/0x720
  worker_thread+0x4b/0x4f0
  kthread+0x10f/0x150
  ? process_one_work+0x720/0x720
  ? kthread_create_on_node+0x40/0x40
  ret_from_fork+0x2e/0x40
 RIP: generic_make_request_checks+0x4d/0x610 RSP: ffffc90001337bb8

In btrfs_dev_replace_finishing(), we will call
btrfs_rm_dev_replace_blocked() to wait bios before destroying the target
device when scrub is finished normally.

However when dev-replace is aborted, either due to error or cancelled by
scrub, we didn't wait for bios, this can lead to use-after-free if there
are bios holding the target device.

Furthermore, for raid56 scrub, at least 2 places are calling
btrfs_map_sblock() without protection of bio_counter, leading to the
problem.

This patch fixes the problem:
1) Wait for bio_counter before freeing target device when canceling
   replace
2) When calling btrfs_map_sblock() for raid56, use bio_counter to
   protect the call.

Cc: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/dev-replace.c
fs/btrfs/raid56.c
fs/btrfs/scrub.c

index de7b2c897fe05ac18cd94974be2c3411670ccf97..32b2ffbb571764023d54ff4292d18a44f027fbbf 100644 (file)
@@ -546,8 +546,10 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
                mutex_unlock(&fs_info->chunk_mutex);
                mutex_unlock(&fs_info->fs_devices->device_list_mutex);
                mutex_unlock(&uuid_mutex);
+               btrfs_rm_dev_replace_blocked(fs_info);
                if (tgt_device)
                        btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
+               btrfs_rm_dev_replace_unblocked(fs_info);
                mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 
                return scrub_ret;
index 7bf65d9a9bd3230616ecbd6e994390cee8cc4b31..d8ea0eb76325e9b25d42dfa4a99c63918981aa2f 100644 (file)
@@ -2198,6 +2198,8 @@ static void read_rebuild_work(struct btrfs_work *work)
 /*
  * The following code is used to scrub/replace the parity stripe
  *
+ * Caller must have already increased bio_counter for getting @bbio.
+ *
  * Note: We need make sure all the pages that add into the scrub/replace
  * raid bio are correct and not be changed during the scrub/replace. That
  * is those pages just hold metadata or file data with checksum.
@@ -2235,6 +2237,12 @@ raid56_parity_alloc_scrub_rbio(struct btrfs_fs_info *fs_info, struct bio *bio,
        ASSERT(rbio->stripe_npages == stripe_nsectors);
        bitmap_copy(rbio->dbitmap, dbitmap, stripe_nsectors);
 
+       /*
+        * We have already increased bio_counter when getting bbio, record it
+        * so we can free it at rbio_orig_end_io().
+        */
+       rbio->generic_bio_cnt = 1;
+
        return rbio;
 }
 
@@ -2677,6 +2685,12 @@ raid56_alloc_missing_rbio(struct btrfs_fs_info *fs_info, struct bio *bio,
                return NULL;
        }
 
+       /*
+        * When we get bbio, we have already increased bio_counter, record it
+        * so we can free it at rbio_orig_end_io()
+        */
+       rbio->generic_bio_cnt = 1;
+
        return rbio;
 }
 
index 26dbe563fed0a698a14093e034d6e92b82c15a22..e9612016325ffd1c337ad58613ef21e8ce12969c 100644 (file)
@@ -2191,6 +2191,7 @@ static void scrub_missing_raid56_pages(struct scrub_block *sblock)
        int ret;
        int i;
 
+       btrfs_bio_counter_inc_blocked(fs_info);
        ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
                        &length, &bbio);
        if (ret || !bbio || !bbio->raid_map)
@@ -2235,6 +2236,7 @@ static void scrub_missing_raid56_pages(struct scrub_block *sblock)
 rbio_out:
        bio_put(bio);
 bbio_out:
+       btrfs_bio_counter_dec(fs_info);
        btrfs_put_bbio(bbio);
        spin_lock(&sctx->stat_lock);
        sctx->stat.malloc_errors++;
@@ -2778,6 +2780,8 @@ static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
                goto out;
 
        length = sparity->logic_end - sparity->logic_start;
+
+       btrfs_bio_counter_inc_blocked(fs_info);
        ret = btrfs_map_sblock(fs_info, BTRFS_MAP_WRITE, sparity->logic_start,
                               &length, &bbio);
        if (ret || !bbio || !bbio->raid_map)
@@ -2805,6 +2809,7 @@ static void scrub_parity_check_and_repair(struct scrub_parity *sparity)
 rbio_out:
        bio_put(bio);
 bbio_out:
+       btrfs_bio_counter_dec(fs_info);
        btrfs_put_bbio(bbio);
        bitmap_or(sparity->ebitmap, sparity->ebitmap, sparity->dbitmap,
                  sparity->nsectors);