Re: BUG: unable to handle kernel NULL pointer dereference at (null)

[Josef Bacik <josef@redhat.com>]

On Tue, Apr 05, 2011 at 09:21:55PM +0200, Johannes Hirte wrote:
> On Tuesday 05 April 2011 20:53:24 Josef Bacik wrote:
> > On Tue, Apr 05, 2011 at 08:52:21PM +0200, Johannes Hirte wrote:
> > > On Tuesday 05 April 2011 19:42:03 Josef Bacik wrote:
> > > > On Tue, Apr 05, 2011 at 07:38:13PM +0200, Johannes Hirte wrote:
> > > > > With the latest btrfs changes, I got this Oops when doing rm on a
> > > > > large directory:
> > > > > 
> > > > > BUG: unable to handle kernel NULL pointer dereference at   (null)
> > > > > IP: [<c101c838>] kunmap+0x46/0x46
> > > > > *pdpt = 0000000034a85001 *pde = 0000000000000000
> > > > > Oops: 0000 [#1] PREEMPT SMP
> > > > > last sysfs file: /sys/devices/virtual/vtconsole/vtcon1/uevent
> > > > > Modules linked in: snd_seq_oss snd_seq_midi_event snd_seq
> > > > > snd_seq_device snd_pcm_oss snd_mixer_oss fuse dm_crypt dm_mod usbhid
> > > > > snd_intel8x0 snd_ac97_codec ac97_bus snd_pcm snd_timer sr_mod cdrom
> > > > > sg snd fschmd e1000 uhci_hcd snd_page_alloc i2c_i801 [last unloaded:
> > > > > microcode]
> > > > > 
> > > > > Pid: 1156, comm: btrfs-transacti Tainted: G        W  
> > > > > 2.6.39-rc1-00262- gc53813f #20 FUJITSU SIEMENS SCENIC P / SCENICO
> > > > > P/D1561
> > > > > EIP: 0060:[<c101c838>] EFLAGS: 00010296 CPU: 1
> > > > > EIP is at kmap+0x0/0x38
> > > > > EAX: 00000000 EBX: 00000000 ECX: 00000000 EDX: 00000010
> > > > > ESI: f5bc6400 EDI: f3c75520 EBP: f3c755f0 ESP: f58f9e10
> > > > > 
> > > > >  DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
> > > > > 
> > > > > Process btrfs-transacti (pid: 1156, ti=f58f8000 task=f6516f40
> > > > > task.ti=f58f8000)
> > > > > 
> > > > > Stack:
> > > > >  c1186d15 ffc22000 f58f9ec0 00000010 f3c75610 00000000 f5885780
> > > > >  f52339e8 00000009 f5bc6400 00010000 00000000 f6415800 f3c75638
> > > > >  000008bb f5bc63c0 f58857b4 f60b68a0 00000040 f52338e8 ffc22000
> > > > >  00000000 00000008 00000010
> > > > > 
> > > > > Call Trace:
> > > > >  [<c1186d15>] ? btrfs_write_out_cache+0x60c/0xa3c
> > > > >  [<c114a815>] ? btrfs_write_dirty_block_groups+0x400/0x494
> > > > >  [<c11566a7>] ? commit_cowonly_roots+0xa9/0x180
> > > > >  [<c1157799>] ? btrfs_commit_transaction+0x2ee/0x59c
> > > > >  [<c1037c85>] ? wake_up_bit+0x16/0x16
> > > > >  [<c1152a83>] ? transaction_kthread+0x149/0x1d6
> > > > >  [<c101d1b9>] ? complete+0x28/0x36
> > > > >  [<c115293a>] ? btrfs_congested_fn+0x5d/0x5d
> > > > >  [<c10379c4>] ? kthread+0x63/0x68
> > > > >  [<c1037961>] ? kthread_worker_fn+0xeb/0xeb
> > > > >  [<c13cba36>] ? kernel_thread_helper+0x6/0xd
> > > > > 
> > > > > Code: 8d 8a 00 e4 54 c1 2b 8a 8c e7 54 c1 81 f9 00 08 00 00 74 11 81
> > > > > f9 00 0c 00 00 75 0e 83 3d 10 2f 60 c1 02 75 05 e9 5e a3 04 00 c3
> > > > > <8b> 10 c1 ea 1e c1 e2 0a 8d 8a 00 e4 54 c1 2b 8a 8c e7 54 c1 81
> > > > > EIP: [<c101c838>] kmap+0x0/0x38 SS:ESP 0068:f58f9e10
> > > > > CR2: 0000000000000000
> > > > > ---[ end trace c8511126ee91dfdf ]---
> > > > > 
> > > > > This is the second Oops. On the first one I wasn't able to catch the
> > > > > backtrace, but IIRC the bug happend on kmap not kunmap the first
> > > > > time.
> > > > 
> > > > Yeah I think I know what this is but I need somebody to verify it for
> > > > me. Can you run with this patch and let me know what happens?  Thanks,
> > > > 
> > > > Josef
> > > > 
> > > > diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
> > > > index 74bc432..5e6f4b3 100644
> > > > --- a/fs/btrfs/free-space-cache.c
> > > > +++ b/fs/btrfs/free-space-cache.c
> > > > @@ -624,6 +624,7 @@ int btrfs_write_out_cache(struct btrfs_root *root,
> > > > 
> > > >  		next_page = false;
> > > > 
> > > > +		BUG_ON(index > last_index);
> > > > 
> > > >  		if (index == 0) {
> > > >  		
> > > >  			start_offset = first_page_offset;
> > > >  			offset = start_offset;
> > > > 
> > > > @@ -732,6 +733,7 @@ int btrfs_write_out_cache(struct btrfs_root *root,
> > > > 
> > > >  		struct btrfs_free_space *entry =
> > > >  		
> > > >  			list_entry(pos, struct btrfs_free_space, list);
> > > > 
> > > > +		BUG_ON(index > last_index);
> > > > 
> > > >  		page = find_get_page(inode->i_mapping, index);
> > > >  		
> > > >  		addr = kmap(page);
> > > 
> > > Hm, I tried but now I hit the
> > > BUG_ON(block_group->total_bitmaps >= max_bitmaps); in add_new_bitmap in
> > > fs/btrfs/free-space-cache.c:1255 when booting the system.
> > 
> > Can you mount -o clear_cache to make sure it's not the cache thats causing
> > that? Thanks,
> > 
> > Josef
> 
> Mounting  with clear_cache under 2.6.38 helped. I was able to boot and test 
> with your patch an hit the second BUG_ON on free-space-cache.c:738.
> 

Perfect can you try this and verify you don't panic anymore please?  Thanks,

Josef

---
 fs/btrfs/free-space-cache.c |   18 ++++++++++++++++++
 1 files changed, 18 insertions(+), 0 deletions(-)

diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 74bc432..33287e8 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -522,6 +522,7 @@ int btrfs_write_out_cache(struct btrfs_root *root,
 	int bitmaps = 0;
 	int ret = 0;
 	bool next_page = false;
+	bool out_of_space = false;
 
 	root = root->fs_info->tree_root;
 
@@ -629,6 +630,11 @@ int btrfs_write_out_cache(struct btrfs_root *root,
 			offset = start_offset;
 		}
 
+		if (index > last_index) {
+			out_of_space = true;
+			break;
+		}
+
 		page = find_get_page(inode->i_mapping, index);
 
 		addr = kmap(page);
@@ -732,6 +738,10 @@ int btrfs_write_out_cache(struct btrfs_root *root,
 		struct btrfs_free_space *entry =
 			list_entry(pos, struct btrfs_free_space, list);
 
+		if (index > last_index) {
+			out_of_space = true;
+			break;
+		}
 		page = find_get_page(inode->i_mapping, index);
 
 		addr = kmap(page);
@@ -754,6 +764,14 @@ int btrfs_write_out_cache(struct btrfs_root *root,
 		index++;
 	}
 
+	if (out_of_space) {
+		ret = 0;
+		unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
+				     i_size_read(inode) - 1, &cached_state,
+				     GFP_NOFS);
+		goto out_free;
+	}
+
 	/* Zero out the rest of the pages just to make sure */
 	while (index <= last_index) {
 		void *addr;
-- 
1.7.2.3