[Bug 228174] [dump] dump(8) can read garbage and loop forever

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=228174

            Bug ID: 228174
           Summary: [dump] dump(8) can read garbage and loop forever
           Product: Base System
           Version: 11.1-STABLE
          Hardware: Any
                OS: Any
            Status: New
          Severity: Affects Some People
          Priority: ---
         Component: bin
          Assignee: stable at FreeBSD.org
          Reporter: eugen at freebsd.org
                CC: fs at FreeBSD.org, longwitz at incore.de,
                    mckusick at FreeBSD.org

Hi!

I have several FreeBSD boxes with UFS2 and soft-updates enabled (no SU+J)
mounted as (ufs, local, soft-updates). They periodically run dump(8) to make
zero-level backups once a week and first level incremental backup on other days
of week.

The command line is like: dump -$level -C 32 -h 0 -au -f - /home
That is, not snapshots used and cache size is big.

Sometimes it finished just fine. Sometimes it its four children start to eat
CPU at max and loop forever not making progress and never finishing until
killed manually.

I've rebuilt /sbin/dump with debugging options -O0 -g and installed unstripped
binary and when the problem reproduced, I ran sysctl
kern.corefile=/var/tmp/%N.%P.core && killall -QUIT dump. So, I have five core
files with debug info.

I've digged it a bit and I think I found the problem. Here is what do I get for
FreeBSD 11.1-STABLE/amd64 r332356 and file system having 41943040 sectors, 512
bytes per sector and standard bsize=32768, fsize=4096:

Core was generated by `dump: /dev/mirror/gm0s1g: pass 4: 60.53% done, finished
in 0:03 at Fri May 11 03'.
Program terminated with signal 3, Quit.
Reading symbols from /lib/libc.so.7...done.
Loaded symbols for /lib/libc.so.7
Reading symbols from /libexec/ld-elf.so.1...done.
Loaded symbols for /libexec/ld-elf.so.1
#0  0x000000080098bfea in _read () from /lib/libc.so.7
(gdb) bt
#0  0x000000080098bfea in _read () from /lib/libc.so.7
#1  0x0000000000408806 in atomic (func=0x4021f4 <read at plt>, fd=5,
buf=0x7ffffffedb44 "", count=4)
    at /usr/local/src/sbin/dump/tape.c:877
#2  0x0000000000407d75 in flushtape () at /usr/local/src/sbin/dump/tape.c:245
#3  0x000000000040814b in dumpblock (blkno=3906360519982919218, size=32768)
    at /usr/local/src/sbin/dump/tape.c:188
#4  0x000000000040cd38 in ufs2_blksout (dp=0x8011a9fc0, blkp=0x7ffffffedc58,
frags=679,
    ino=522029, last=1) at /usr/local/src/sbin/dump/traverse.c:704
#5  0x000000000040cfea in dmpindir (dp=0x8011a9fc0, ino=522029, blk=5832576,
ind_level=0,
    size=0x7fffffffdd58) at /usr/local/src/sbin/dump/traverse.c:609
#6  0x000000000040c322 in dumpino (dp=0x8011a9fc0, ino=522029)
    at /usr/local/src/sbin/dump/traverse.c:573
#7  0x0000000000404cd1 in main (argc=0, argv=0x7fffffffec60)
    at /usr/local/src/sbin/dump/main.c:579

Note insanely large blkno for frame 3.
I think the problem is in the dmpindir() function, frame 5.

Note ino=522029. It corresponds to some large text CSV file this box produces
periodically and "find /home -inode 522029 -ls" shows its pathname and I can
see its contents to the end with "less" command just fine.

However:

(gdb) frame 5
#5  0x000000000040cfea in dmpindir (dp=0x8011a9fc0, ino=522029, blk=5832576,
ind_level=0,
    size=0x7fffffffdd58) at /usr/local/src/sbin/dump/traverse.c:609
(gdb) l dmpindir
580      * Read indirect blocks, and pass the data blocks to be dumped.
581      */
582     static void
583     dmpindir(union dinode *dp, ino_t ino, ufs2_daddr_t blk, int ind_level,
584             off_t *size)
585     {
586             union {
587                     ufs1_daddr_t ufs1[MAXBSIZE / sizeof(ufs1_daddr_t)];
588                     ufs2_daddr_t ufs2[MAXBSIZE / sizeof(ufs2_daddr_t)];
589             } idblk;
590             int i, cnt, last;
591
592             if (blk != 0)
593                     bread(fsbtodb(sblock, blk), (char *)&idblk,
594                         (int)sblock->fs_bsize);
595             else
596                     memset(&idblk, 0, sblock->fs_bsize);
597             if (ind_level <= 0) {
598                     if (*size > NINDIR(sblock) * sblock->fs_bsize) {
599                             cnt = NINDIR(sblock) * sblock->fs_frag;
(gdb) p blk
$1 = 5832576
(gdb) p idblk.ufs2[0]
$2 = 3329910757767148590

The code bread's garbage from the file system (line 593) and uses it later
without any sanity checks.

-- 
You are receiving this mail because:
You are the assignee for the bug.