Re: fsck segfaults on rpi3 running 13-stable (and on 14-CURRENT analyzing the same file system that resulted from the 13-STABLE crash)

John-Mark Gurney wrote this message on Mon, Feb 13, 2023 at 15:25 -0800:

After analyzing the remaining crashes that were detected, the following
was found..  This was done w/ a minimal FS,  It is possible that it'll
find more crashes w/ a slightly more complex FS, e.g. one w/ indirect
blocks, and other file types.

Note that FS's that exhibit these are available at:
https://www.funkthat.com/~jmg/FreeBSD/ffs.afl/

---- FIRST ----
std_checkblkavail doesn't check that blkno is within valid range, e.g.
if blkno is negative, it'll segfault:
0x0000000000121d6c in std_checkblkavail (blkno=blkno@entry=-288230376151711688, frags=frags@entry=1) at /usr/src/sbin/fsck_ffs/fsutil.c:1149
1149                    if (testbmap(blkno + j))


---- SECOND ----
pass5 doesn't check that cg_nextfreeoff is sane/within range, this
causes a large value to be passed to memset, in this case:
#1  0x000000000012fee4 in pass5 () at /usr/src/sbin/fsck_ffs/pass5.c:241
241                     memset(cg_inosused(newcg), 0, (size_t)(mapsize));
(gdb) print *newcg
$7 = {cg_firstfield = 0, cg_magic = 590421, cg_old_time = 0, cg_cgx = 0,
  cg_old_ncyl = 0, cg_old_niblk = 0, cg_ndblk = 256, cg_cs = {cs_ndir = 0,
  cs_nbfree = 0, cs_nifree = 128, cs_nffree = 0}, cg_rotor = 0,
  cg_frotor = 0, cg_irotor = 0, cg_frsum = {0, 0, 0, 0, 0, 0, 0, 0},
  cg_old_btotoff = 0, cg_old_boff = 0, cg_iusedoff = 168,
  cg_freeoff = 184, cg_nextfreeoff = 61341980,
  cg_clustersumoff = 54526164, cg_clusteroff = 54526232,
  cg_nclusterblks = 32, cg_niblk = 128, cg_initediblk = 128,
  cg_unrefs = 0, cg_sparecon32 = {0}, cg_ckhash = 3548071837,
  cg_time = 1676327229, cg_sparecon64 = {0, 0, 0}, cg_space = ""}

and mapsize is:
(gdb) print newcg->cg_nextfreeoff - newcg->cg_iusedoff
$6 = 61341812

which overflows buf, which is MAXBSIZE, or 8k.

---- THIRD ----
allocino doesn't make sure that cg_iusedoff is sane.  In this case,
cg_iusedoff is 4294965672, which means that in allocino, the setbit
function call will access invalid memory.

0x0000000000127168 in allocino (request=2, type=16877) at /usr/src/sbin/fsck_ffs/inode.c:1379
1379            setbit(cg_inosused(cgp), ino % sblock.fs_ipg);

(gdb) print *cgp
$14 = {cg_firstfield = 0, cg_magic = 590421, cg_old_time = 0, 
  cg_cgx = 0, cg_old_ncyl = 0, cg_old_niblk = 0, cg_ndblk = 256, 
  cg_cs = {cs_ndir = 2, cs_nbfree = 23, cs_nifree = 124, 
    cs_nffree = 21}, cg_rotor = 0, cg_frotor = 0, cg_irotor = 0, 
  cg_frsum = {0, 0, 0, 0, 0, 0, 0, 3}, cg_old_btotoff = 0, 
  cg_old_boff = 0, cg_iusedoff = 4294965672, cg_freeoff = 183, 
  cg_nextfreeoff = 284, cg_clustersumoff = 212, cg_clusteroff = 280, 
  cg_nclusterblks = 32, cg_niblk = 128, cg_initediblk = 128, 
  cg_unrefs = 0, cg_sparecon32 = {0}, cg_ckhash = 3548071837, 
  cg_time = 1676327229, cg_sparecon64 = {0, 0, 0}, cg_space = "\017"}

-- 
  John-Mark Gurney				Voice: +1 415 225 5579

     "All that I will do, has been done, All that I have, has not."