Add bounds checking to zil_parse #16308
Conversation
|
Added a space to meet code style check. |
|
Modified commit message to meet format requirements. |
There was a problem hiding this comment.
Generally it looks good to me, except there is no EUCLEAR code on FreeBSD, and I am not sure how applicable it is to this case on Linux. When zil_read_log_block() does similar checks, it returns ECKSUM.
Another thought: ZFS relies on checksums to check data correct, and if they match, there should be endless number of ways to crash the system with maliciously crafted pool. I am not saying this patch is bad, but it is a drop in the ocean.
There was a problem hiding this comment.
One more thought: previously ASSERT()'s there would crash the system if built with debug, that could be used for testing quiet, while error return in this case may pass unnoticed. I am actually surprised that your pool import failed on testing, since at least checksum errors are normal completion for ZIL replay and is not fatal. Though may be it is only specific for ECKSUM.
|
Thanks for your time. @amotin
Oh sorry, I'm not familiar with FreeBSD. I also consider EIO and ENOSPC before. I refer to similar patches in Linux, so I use UCLEAN here. What is your suggestion about the return code? Thanks. I think these are two scenarios:
From a security perspective, this is an attack surface no matter how many such crashes there are. So we need to add sanity checks.
Yes, I also noticed this issue before when writing the patch. There is not much useful information in the error message. So should we maybe keep ASSERT()?
I see, zil_check_log_chain thinks ECKSUM is not a fatal error. But if it's a crash issue I think we need to get a fatal error. |
I'd use ECKSUM to make it non-error. It may be not very descriptive, but there is no much good in aborting pool import due to corrupted ZIL.
May be. |
Make sure log record don't stray beyond valid memory region.
There is a lack of verification of the space occupied by fixed members
of lr_t in the zil_parse.
We can create a crafted image to trigger an out of bounds read by
following these steps:
1) Do some file operations and reboot to simulate abnormal exit
without umount
2) zil_chain.zc_nused: 0x1000
3) First lr_t
lr_t.lrc_txtype: 0x0
lr_t.lrc_reclen: 0x1000-0xb8-0x1
lr_t.lrc_txg: 0x0
lr_t.lrc_seq: 0x1
4) Update checksum in zil_chain.zc_eck
Fix:
Add some checks to make sure the remaining bytes are large enough to
hold an log record.
Signed-off-by: XDTG <[email protected]>
Add bounds checking to zil_parse to ensure log records don't stray beyond valid memory region.
Motivation and Context
This change fixes the memory out of bounds issue mentioned in #16246.
Description
Before accessing the next log record header, check that there are enough bytes left to accommodate a log record. And ensure the
lr->lrc_reclenis appropriate.How Has This Been Tested?
An error is returned when using the crafted image mentioned in the #16246.
Types of changes
Checklist:
Signed-off-by.