Hi Tytso,
I find a error in ext2fs_extent_delete when node has no entries left.
A ext4 filesystem have a file whose extent looks like this:
debugfs: ex <950>
Level Entries Logical Physical Length Flags
0/ 1 1/ 1 102 - 102 32910 1
1/ 1 1/ 6 102 - 104 196802 - 196804 3
1/ 1 2/ 6 105 - 140 230934 - 230969 36
1/ 1 3/ 6 154 - 159 199322 - 199327 6
1/ 1 4/ 6 393 - 405 197123 - 197135 13
1/ 1 5/ 6 541 - 553 230439 - 230451 13
1/ 1 6/ 6 1058 - 1088 196640 - 196670 31
In the program, call scan_extent_node recursively to find the leaf entry
of level 1,
the variable "problem" will be assigned PR_1_EXTENT_END_OUT_OF_BOUNDS
because the extent of level 0 has a problem with the range.
so the program thinks this extent is wronged and call
ext2fs_extent_delete to delete it.
But when the entry is the last entry of level 1,
the program will run handle->level-- get parent entry and delete it in
ext2fs_extent_delete.
However scan_extent_node is a recursive call, after the function returns,
handle->level is not restored to the value of the current
level(handle->level is equal 0 new).
So calling ext2fs_extent_get(ehandle, EXT2_EXTENT_UP, &extent) again
will return EXT2_ET_EXTENT_NO_UP,
and then print failed.
Command output:
[root@localhost ~]# e2fsck -a ram1gdb
ram1gdb: recovering journal
ram1gdb contains a file system with errors, check forced.
ram1gdb: Inode 950, end of extent exceeds allowed value
(logical block 102, physical block 196802, len 3)
CLEARED.
ram1gdb: Inode 950, end of extent exceeds allowed value
(logical block 105, physical block 230934, len 36)
CLEARED.
ram1gdb: Inode 950, end of extent exceeds allowed value
(logical block 154, physical block 199322, len 6)
CLEARED.
ram1gdb: Inode 950, end of extent exceeds allowed value
(logical block 393, physical block 197123, len 13)
CLEARED.
ram1gdb: Inode 950, end of extent exceeds allowed value
(logical block 541, physical block 230439, len 13)
CLEARED.
ram1gdb: Inode 950, end of extent exceeds allowed value
(logical block 1058, physical block 196640, len 31)
CLEARED.
ram1gdb: Failed to iterate extents in inode 950
(op EXT2_EXTENT_UP, blk 196640, lblk 1058): No 'up' extent
ram1gdb: UNEXPECTED INCONSISTENCY; RUN fsck MANUALLY.
(i.e., without -a or -p options)
/etc/e2fsck.conf:
[root@localhost ~]# cat /etc/e2fsck.conf
[problems]
...
0x01006e = {
preen_ok = true
}
...
gdb:
(gdb) p *handle
$17 = {magic = 2133571346, fs = 0x5555555c7430, ino = 950, inode =
0x5555555bd6b0, inodebuf = {i_mode = 0, i_uid = 0, i_size = 0, i_atime =
0, i_ctime = 0, i_mtime = 0, i_dtime = 0, i_gid = 0,
i_links_count = 0, i_blocks = 0, i_flags = 0, osd1 = {linux1 =
{l_i_version = 0}, hurd1 = {h_i_translator = 0}}, i_block = {0 <repeats
15 times>}, i_generation = 0, i_file_acl = 0, i_size_high = 0,
i_faddr = 0, osd2 = {linux2 = {l_i_blocks_hi = 0, l_i_file_acl_high = 0,
l_i_uid_high = 0, l_i_gid_high = 0, l_i_checksum_lo = 0, l_i_reserved =
0}, hurd2 = {h_i_frag = 0 '\000', h_i_fsize = 0 '\000',
h_i_mode_high = 0, h_i_uid_high = 0, h_i_gid_high = 0, h_i_author =
0}}}, type = 62218, level = 0, max_depth = 1, max_paths = 2, path =
0x5555555db8d0}
(gdb) bt
#0 ext2fs_extent_delete (handle=0x5555555db7d0, flags=0) at extent.c:1655
#1 0x000055555556c5f9 in scan_extent_node (ctx=0x5555555ab420, pctx=,
pb=, start_block=102, end_block=102, eof_block=102,
ehandle=0x5555555db7d0, try_repairs=1)
at pass1.c:2965
#2 0x000055555556c513 in scan_extent_node (ctx=0x5555555ab420, pctx=,
pb=, start_block=0, end_block=0, eof_block=102, ehandle=0x5555555db7d0,
try_repairs=1) at pass1.c:3048
#3 0x000055555556f0bf in check_blocks_extents (pb=0x7fffffffdbb0,
pctx=0x7fffffffddf0, ctx=0x5555555ab420) at pass1.c:3279
#4 check_blocks (ctx=0x5555555ab420, pctx=0x7fffffffddf0,
block_buf=0x5555555b4980 "", ea_ibody_quota=0x7fffffffdd90) at pass1.c:3402
#5 0x0000555555571693 in e2fsck_pass1 (ctx=0x5555555ab420) at pass1.c:1969
#6 0x000055555556921a in e2fsck_run (ctx=0x5555555ab420) at e2fsck.c:262
#7 0x0000555555564d6e in main (argc=, argv=) at unix.c:1922
regards,
Zhan Chengbin
