Re: [PATCH 1/3] coredump: Fixes core_pipe_limit sysctl proc_handler

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 




On 11/14/24 02:34, Lin Feng wrote:
Hi,

On 11/13/24 22:15, Nicolas Bouchinet wrote:
Hi Lin,

Thanks for your review.

On 11/13/24 03:35, Lin Feng wrote:
Hi,

see comments below please.

On 11/12/24 21:13, nicolas.bouchinet@xxxxxxxxxxx wrote:
From: Nicolas Bouchinet <nicolas.bouchinet@xxxxxxxxxxx>

proc_dointvec converts a string to a vector of signed int, which is
stored in the unsigned int .data core_pipe_limit.
It was thus authorized to write a negative value to core_pipe_limit
sysctl which once stored in core_pipe_limit, leads to the signed int
dump_count check against core_pipe_limit never be true. The same can be
achieved with core_pipe_limit set to INT_MAX.

Any negative write or >= to INT_MAX in core_pipe_limit sysctl would
hypothetically allow a user to create very high load on the system by
running processes that produces a coredump in case the core_pattern
sysctl is configured to pipe core files to user space helper.
Memory or PID exhaustion should happen before but it anyway breaks the
core_pipe_limit semantic

This commit fixes this by changing core_pipe_limit sysctl's proc_handler
to proc_dointvec_minmax and bound checking between SYSCTL_ZERO and
SYSCTL_INT_MAX.

Fixes: a293980c2e26 ("exec: let do_coredump() limit the number of concurrent dumps to pipes")
Signed-off-by: Nicolas Bouchinet <nicolas.bouchinet@xxxxxxxxxxx>
---
   fs/coredump.c | 7 +++++--
   1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/fs/coredump.c b/fs/coredump.c
index 7f12ff6ad1d3e..8ea5896e518dd 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -616,7 +616,8 @@ void do_coredump(const kernel_siginfo_t *siginfo)
   		cprm.limit = RLIM_INFINITY;
dump_count = atomic_inc_return(&core_dump_count);
-		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
+		if ((core_pipe_limit && (core_pipe_limit < dump_count)) ||
+		    (core_pipe_limit && dump_count == INT_MAX)) {
While comparing between 'unsigned int' and 'signed int', C deems them both
to 'unsigned int', so as an insane user sets core_pipe_limit to INT_MAX,
and dump_count(signed int) does overflow INT_MAX, checking for
'core_pipe_limit < dump_count' is passed, thus codes skips core dump.

So IMO it's enough after changing proc_handler to proc_dointvec_minmax.
Indeed, but the dump_count == INT_MAX is not here to catch overflow but
if both dump_count
and core_pipe_limit are equal to INT_MAX. core_pipe_limit will not be
inferior to dump_count.
Or maybe I am missing something ?

Extracted from man core:
        Since Linux 2.6.32, the /proc/sys/kernel/core_pipe_limit can be used to
        defend against this possibility.  The value in this  file  defines  how
        many  concurrent crashing processes may be piped to user-space programs
        in parallel.  If this value is exceeded, then those crashing  processes
        above  this  value are noted in the kernel log and their core dumps are
        skipped.

Since no spinlock protecting us, due to the concurrent running of
atomic_inc_return(&core_dump_count), even with the changing above
it's not guaranteed that core_dump_count can't exceed core_pipe_limit).
As you said, suppose both of them are equal to INT_MAX(0x7fffffff),
and before any dummping thread drops core_dump_count, one new thread
comes in then hits atomic_inc_return(&core_dump_count) and now
(unsigned int)core_dump_count is 0x80000000, but original codes checking
for core_pipe_limit still works as expected.

You are absolutely right about this. Moreover, as stated in my commit
message, pid or memory exhaustion should occur before reaching this
scenario.

Thank's for your comment and review. I'll fix and push a v2.


Please correct me if I'm wrong :)

Thanks,
linfeng






[Index of Archives]     [Kernel Newbies]     [Security]     [Netfilter]     [Bugtraq]     [Linux PPP]     [Linux FS]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Samba]     [Video 4 Linux]     [Linmodem]     [Device Mapper]     [Linux Kernel for ARM]

  Powered by Linux