There is a possibility that 'i_ino' overflows
Hi,
In btrfs, inode number is increased each time a new file or directory is made.
Therefore, if the making deletion of the file is repeated, value of 'i_ino' increases rapidly.
For example, inode number changes as follows.
$ touch foo $ ls -i foo 266 foo $ rm foo $ touch bar $ ls -i bar 267 bar $
And then, length of 'i_ino' and 'objectid' is as follows on the x86 system.
unsigned long i_ino == 32bits
u64 objectid == 64bits
Therefore, in the operation to substitute 'objectid' to 'i_ino', 'i_ino' overflows when 'objectid' 4294967296 is substituted to 'i_ino'. Then, the file with inode number 0 is made.
As a result, ls command has looped infinitely, and btrfsck detected the error.
Please see below.
$ uname -a
Linux luna 2.6.37-rc5 #1 SMP Thu Dec 9 13:02:41 JST 2010 i686 i686 i386 GNU/Linux
$ df -T /test1
Filesystem Type 1K-blocks Used Available Use% Mounted on
/dev/sdd14 btrfs 4162560 56 3717632 1% /test1
$ strace -FfTttx ls /test1
14:03:10.115440 execve("/bin/ls", ["ls", "/test1"], [/* 28 vars */]) = 0 <0.000181>
...
...
14:03:10.123431 stat("/test1", {st_mode=S_IFDIR|0777, st_size=8, ...}) = 0 <0.000017>
14:03:10.123521 open("/test1", O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC) = 3 <0.000018>
14:03:10.123578 fcntl(3, F_GETFD) = 0x1 (flags FD_CLOEXEC) <0.000013>
14:03:10.123637 getdents(3, /* 3 entries */, 32768) = 72 <0.000025>
14:03:10.123712 getdents(3, /* 1 entries */, 32768) = 24 <0.000016>
14:03:10.123768 getdents(3, /* 1 entries */, 32768) = 24 <0.000015>
14:03:10.123824 getdents(3, /* 1 entries */, 32768) = 24 <0.000015>
14:03:10.123880 getdents(3, /* 1 entries */, 32768) = 24 <0.000015>
14:03:10.123936 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
14:03:10.123992 getdents(3, /* 1 entries */, 32768) = 24 <0.000015>
14:03:10.124047 getdents(3, /* 1 entries */, 32768) = 24 <0.000015>
14:03:10.124103 getdents(3, /* 1 entries */, 32768) = 24 <0.000015>
14:03:10.124261 getdents(3, /* 1 entries */, 32768) = 24 <0.000016>
14:03:10.124320 getdents(3, /* 1 entries */, 32768) = 24 <0.000015>
14:03:10.124381 getdents(3, /* 1 entries */, 32768) = 24 <0.000015>
14:03:10.124437 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
14:03:10.124493 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
14:03:10.124549 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
14:03:10.124605 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
14:03:10.124661 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
14:03:10.124717 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
14:03:10.124773 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
14:03:10.124840 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
14:03:10.124896 getdents(3, /* 1 entries */, 32768) = 24 <0.000014>
...
...
$ cd /test1
$ ls -i A123
ls: cannot access A123: No such file or directory
$ touch aaaa
$ ls -i aaaa
1 aaaa
$ # umount /test1
# btrfsck /dev/sdd14
root 5 inode 0 errors 2001
unresolved ref dir 256 index 4294967041 namelen 4 name A123 filetype 1 error 4
root 5 inode 1 errors 2001
unresolved ref dir 256 index 4294967042 namelen 4 name aaaa filetype 1 error 4
root 5 inode 4294967296 errors 2000
unresolved ref dir 256 index 4294967041 namelen 4 name A123 filetype 0 error 3
root 5 inode 4294967297 errors 2000
unresolved ref dir 256 index 4294967042 namelen 4 name aaaa filetype 0 error 3
found 28672 bytes used err is 1
total csum bytes: 0
total tree bytes: 28672
total fs tree bytes: 8192
btree space waste bytes: 23191
file data blocks allocated: 0
referenced 0
Btrfs v0.19-36-g70c6c10
#Regards,
Itoh
Quoted 21 lines · Tsutomu Itoh
Hi,
In btrfs, inode number is increased each time a new file or directory is made.
Therefore, if the making deletion of the file is repeated, value of 'i_ino' increases rapidly.For example, inode number changes as follows.
$ touch foo $ ls -i foo 266 foo $ rm foo $ touch bar $ ls -i bar 267 bar $And then, length of '
i_ino' and 'objectid' is as follows on the x86 system.
Quoted 6 lines
unsigned long
i_ino== 32bits
u64 objectid == 64bitsTherefore, in the operation to substitute 'objectid' to '
i_ino', 'i_ino' overflows when 'objectid' 4294967296 is substituted to 'i_ino'. Then, the file with inode number 0 is made.
I think that it is better to recycle inode number that became unused. And, at least, I think that it should make the filesystem not become an abnormal condition.
This patch is a patch that makes an error when inode number is bigger than BTRFS_LAST_FREE_OBJECTID.
--- inode.c | 4 ++++ 1 file changed, 4 insertions(+)
fs/btrfs/inode.c 2010-12-20 09:04:18.000000000 +0900 renamed from fs/btrfs/inode.c 2010-12-16 10:24:48.000000000 +0900 +4 -0
@@ -4529,6 +4529,10 @@ static struct inode *btrfs_new_inode(str inode_init_owner(inode, dir, mode); inode->i_ino = objectid; + if (unlikely(inode->i_ino > (unsigned long)BTRFS_LAST_FREE_OBJECTID)) { + ret = -ENOSPC; + goto fail; + } inode_set_bytes(inode, 0); inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
Quoted 71 lines
As a result, ls command has looped infinitely, and btrfsck detected the error.
Please see below.$ uname -a Linux luna 2.6.37-rc5 #1 SMP Thu Dec 9 13:02:41 JST 2010 i686 i686 i386 GNU/Linux $ df -T /test1 Filesystem Type 1K-blocks Used Available Use% Mounted on /dev/sdd14 btrfs 4162560 56 3717632 1% /test1 $ strace -FfTttx ls /test1 14:03:10.115440 execve("/bin/ls", ["ls", "/test1"], [/* 28 vars */]) = 0 <0.000181> ... ... 14:03:10.123431 stat("/test1", {st_mode=S_IFDIR|0777, st_size=8, ...}) = 0 <0.000017> 14:03:10.123521 open("/test1", O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC) = 3 <0.000018> 14:03:10.123578 fcntl(3, F_GETFD) = 0x1 (flags FD_CLOEXEC) <0.000013> 14:03:10.123637 getdents(3, /* 3 entries */, 32768) = 72 <0.000025> 14:03:10.123712 getdents(3, /* 1 entries */, 32768) = 24 <0.000016> 14:03:10.123768 getdents(3, /* 1 entries */, 32768) = 24 <0.000015> 14:03:10.123824 getdents(3, /* 1 entries */, 32768) = 24 <0.000015> 14:03:10.123880 getdents(3, /* 1 entries */, 32768) = 24 <0.000015> 14:03:10.123936 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> 14:03:10.123992 getdents(3, /* 1 entries */, 32768) = 24 <0.000015> 14:03:10.124047 getdents(3, /* 1 entries */, 32768) = 24 <0.000015> 14:03:10.124103 getdents(3, /* 1 entries */, 32768) = 24 <0.000015> 14:03:10.124261 getdents(3, /* 1 entries */, 32768) = 24 <0.000016> 14:03:10.124320 getdents(3, /* 1 entries */, 32768) = 24 <0.000015> 14:03:10.124381 getdents(3, /* 1 entries */, 32768) = 24 <0.000015> 14:03:10.124437 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> 14:03:10.124493 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> 14:03:10.124549 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> 14:03:10.124605 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> 14:03:10.124661 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> 14:03:10.124717 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> 14:03:10.124773 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> 14:03:10.124840 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> 14:03:10.124896 getdents(3, /* 1 entries */, 32768) = 24 <0.000014> ... ... $ cd /test1 $ ls -i A123 ls: cannot access A123: No such file or directory $ touch aaaa $ ls -i aaaa 1 aaaa $# umount /test1 # btrfsck /dev/sdd14 root 5 inode 0 errors 2001 unresolved ref dir 256 index 4294967041 namelen 4 name A123 filetype 1 error 4 root 5 inode 1 errors 2001 unresolved ref dir 256 index 4294967042 namelen 4 name aaaa filetype 1 error 4 root 5 inode 4294967296 errors 2000 unresolved ref dir 256 index 4294967041 namelen 4 name A123 filetype 0 error 3 root 5 inode 4294967297 errors 2000 unresolved ref dir 256 index 4294967042 namelen 4 name aaaa filetype 0 error 3 found 28672 bytes used err is 1 total csum bytes: 0 total tree bytes: 28672 total fs tree bytes: 8192 btree space waste bytes: 23191 file data blocks allocated: 0 referenced 0 Btrfs v0.19-36-g70c6c10 #Regards,
Itoh
Excerpts from Tsutomu Itoh's message of 2010-12-19 21:02:59 -0500:
Quoted 30 lines
Quoted 21 lines · Tsutomu Itoh
Hi,
In btrfs, inode number is increased each time a new file or directory is made.
Therefore, if the making deletion of the file is repeated, value of 'i_ino' increases rapidly.For example, inode number changes as follows.
$ touch foo $ ls -i foo 266 foo $ rm foo $ touch bar $ ls -i bar 267 bar $And then, length of '
i_ino' and 'objectid' is as follows on the x86 system.Quoted 6 lines
unsigned long
i_ino== 32bits
u64 objectid == 64bitsTherefore, in the operation to substitute 'objectid' to '
i_ino', 'i_ino' overflows when 'objectid' 4294967296 is substituted to 'i_ino'. Then, the file with inode number 0 is made.
Quoted 6 lines
I think that it is better to recycle inode number that became unused. And, at least, I think that it should make the filesystem not become an abnormal condition.
This patch is a patch that makes an error when inode number is bigger than
BTRFS_LAST_FREE_OBJECTID.
Thanks, I've folded this one in. Long term the plan is to do something similar to Josef's block group caching code but for inode numbers. Basically we'll cache the free ones, but it'll be less complex than the block group caching by far.
-chris