User-triggerable WARNING with fuse
Using the attached program I can trigger the following WARNING reliably as a normal user. This happens at least both in 3.8-rc6 and 3.7.5.
The kernel is tainted by proprietary NVIDIA module, but I don't thinks it matters in this case.
[ 4390.882323] ------------[ cut here ]------------ [ 4390.882342] WARNING: at fs/inode.c:280 drop_nlink+0x41/0x50() [ 4390.882345] Hardware name: M50Vm [ 4390.882347] Modules linked in: fuse rfcomm bnep snd_hda_codec_hdmi
nvidia(PO) snd_hda_codec_realtek snd_hda_intel iwldvm snd_hda_codec snd_hwdep uvcvideo btusb videobuf2_vmalloc videobuf2_memops videobuf2_core mac80211 bluetooth snd_seq snd_seq_device snd_pcm asus_laptop iwlwifi cfg80211 snd_page_alloc snd_timer videodev r8169 input_polldev sparse_keymap snd soundcore kvm_intel kvm microcode rfkill i2c_core mii pcspkr uinput hid_generic firewire_ohci sdhci_pci sdhci mmc_core firewire_core crc_itu_t
[ 4390.882438] Pid: 6040, comm: rm Tainted: P W O 3.8.0-rc6 #1 [ 4390.882442] Call Trace: [ 4390.882453] [<ffffffff81032eca>] warn_slowpath_common+0x7a/0xb0 [ 4390.882460] [<ffffffff81032f15>] warn_slowpath_null+0x15/0x20 [ 4390.882466] [<ffffffff81101dd1>] drop_nlink+0x41/0x50 [ 4390.882478] [<ffffffffa00d261f>] fuse_unlink+0xdf/0x130 [fuse] [ 4390.882486] [<ffffffff810f6cfd>] vfs_unlink+0x8d/0x100 [ 4390.882493] [<ffffffff810f6f09>] do_unlinkat+0x199/0x220 [ 4390.882501] [<ffffffff81092e82>] ? call_rcu_sched+0x12/0x20 [ 4390.882508] [<ffffffff81057c2a>] ? __put_cred+0x3a/0x50 [ 4390.882516] [<ffffffff810e9977>] ? sys_faccessat+0x137/0x1e0 [ 4390.882524] [<ffffffff810f94b6>] sys_unlinkat+0x16/0x40 [ 4390.882532] [<ffffffff81490c16>] system_call_fastpath+0x1a/0x1f [ 4390.882536] ---[ end trace 372eef394febff4c ]---
/*
* $ gcc warn.c -o warn -D_FILE_OFFSET_BITS=64 -DFUSE_USE_VERSION=26 -lfuse *
* $ mkdir mnt
* $ ./warn mnt
* $ rm mnt/foo
*/
#include <fuse/fuse.h>
static int readdir_op(const char *path, void *buf,
fuse_fill_dir_t filler, off_t offset, struct fuse_file_info *fi)
{
struct stat st = {.st_mode = S_IFREG | 0666};
filler(buf, "foo", &st, 0);
return 0;
}static int getattr_op(const char *path, struct stat *st)
{
if (strcmp(path, "/") == 0)
st->st_mode = S_IFDIR | 0777;
else
st->st_mode = S_IFREG | 0666;
return 0;
}static int unlink_op(const char *path)
{
return 0;
}static struct fuse_operations ops = {
.readdir = readdir_op,
.getattr = getattr_op,
.unlink = unlink_op,
};
int main(int argc, char *argv[])
{
return fuse_main(argc, argv, &ops, NULL);
}Miklos Szeredi
maintainerQuoted 8 lines · Tero Roponen
Using the attached program I can trigger the following WARNING reliably as a normal user. This happens at least both in 3.8-rc6 and 3.7.5.
The kernel is tainted by proprietary NVIDIA module, but I don't thinks it matters in this case.
Thanks, the attached patch should fix it.
Miklos
From: Miklos Szeredi <[email protected]>
Subject: fuse: don't WARN when nlink is zero
drop_nlink() warns if nlink is already zero. This is triggerable by a buggy userspace filesystem. The cure, I think, is worse than the disease so disable the warning.
--- fs/fuse/dir.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
fs/fuse/dir.c +8 -1
@@ -705,7 +705,14 @@ static int fuse_unlink(struct inode *dirspin_lock (&fc->lock); fi->attr_version = ++fc->attr_version; -drop_nlink (inode); +/* +* If i_nlink == 0 then unlink doesn't make sense, yet this can +* happen if userspace filesystem is careless. It would be +* difficult to enforce correct nlink usage so just ignore this +* condition here +*/ +if (inode->i_nlink >0 ) +drop_nlink (inode);spin_unlock (&fc->lock);fuse_invalidate_attr (inode);fuse_invalidate_attr (dir);
Quoted 11 lines · Miklos Szeredi
Quoted 8 lines · Tero Roponen
Using the attached program I can trigger the following WARNING reliably as a normal user. This happens at least both in 3.8-rc6 and 3.7.5.
The kernel is tainted by proprietary NVIDIA module, but I don't thinks it matters in this case.
Thanks, the attached patch should fix it.
Yes, that fixed it.
Signature · 1 line
Tero Roponen