Thanks for the hint, this is indeed a ZFS bug.
Reported to upstream at https://github.com/openzfs/zfs/pull/15344/
2023-09-25 0:38 GMT+08:00, Ben Hutchings <b...@decadent.org.uk>:
> Control: reassign -1 src:zfs-linux
>
> On Fri, 2023-09-22 at 16:13 +0000, WHR wrote:
>> Package: src:linux
>> Version: 6.5.3-1
>> Severity: normal
>> X-Debbugs-Cc: msl0000023...@gmail.com, msl0000023...@gmail.com
>>
>>
>> Taking executable file /usr/bin/ssh to demonstrate the issue:
>>
>> # which ssh
>> /usr/bin/ssh
>> # ssh
>>
>> usage: ssh [-46AaCfGgKkMNnqsTtVvXxYy] [-B bind_interface]
>> [-b bind_address] [-c cipher_spec] [-D [bind_address:]port]
>> [-E log_file] [-e escape_char] [-F configfile] [-I pkcs11]
>> [-i identity_file] [-J [user@]host[:port]] [-L address]
>> [-l login_name] [-m mac_spec] [-O ctl_cmd] [-o option] [-p
>> port]
>> [-Q query_option] [-R address] [-S ctl_path] [-W host:port]
>> [-w local_tun[:remote_tun]] destination [command]
>> # chattr +i /usr/bin/ssh
>>
>> # ssh
>> Segmentation fault
>>
>>
>> By trying to load the program via ld.so(1) with truss (actually strace),
>> it
>> shows that a mmap(2) call used to load the data segument failed due to
>> EPERM:
>>
>> # truss -s 128 -f /lib/powerpc64-linux-gnu/ld64.so.1 /usr/bin/ssh
>> execve("/lib/powerpc64-linux-gnu/ld64.so.1",
>> ["/lib/powerpc64-linux-gnu/ld64.so.1", "/usr/bin/ssh"], 0x7fffc0380530 /*
>> 29 vars */) = 0
>> brk(NULL) = 0x1000db60000
>> openat(AT_FDCWD, "/usr/bin/ssh", O_RDONLY|O_CLOEXEC) = 3
>> read(3,
>> "\177ELF\2\2\1\0\0\0\0\0\0\0\0\0\0\3\0\25\0\0\0\1\0\0\0\0\0\22h\220\0\0\0\0\0\0\0@\0\0\0\0\0\22\4\330\0\0\0\1\0@\08\0\t\0@\0\35\0\34\0\0\0\6\0\0\0\4\0\0\0\0\0\0\0@\0\0\0\0\0\0\0@\0\0\0\0\0\0\0@\0\0\0\0\0\0\1\370\0\0\0\0\0\0\1\370\0\0\0\0\0\0\0\10\0\0\0\3\0\0\0\4"...,
>> 832) = 832
>> mmap(NULL, 1259760, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 3,
>> 0)
>> = 0x7fff93720000
>> mprotect(0x7fff93830000, 65536, PROT_NONE) = 0
>> mmap(0x7fff93840000, 131072, PROT_READ|PROT_WRITE,
>> MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0x110000) = -1 EPERM (Operation
>> not permitted)
>> close(3) = 0
>> writev(2, [{iov_base="/usr/bin/ssh", iov_len=12}, {iov_base=": ",
>> iov_len=2}, {iov_base="error while loading shared libraries", iov_len=36},
>> {iov_base=": ", iov_len=2}, {iov_base="/usr/bin/ssh", iov_len=12},
>> {iov_base=": ", iov_len=2}, {iov_base="failed to map segment from shared
>> object", iov_len=40}, {iov_base="", iov_len=0}, {iov_base="", iov_len=0},
>> {iov_base="\n", iov_len=1}], 10/usr/bin/ssh: error while loading shared
>> libraries: /usr/bin/ssh: failed to map segment from shared object
>> ) = 107
>> exit_group(127) = ?
>> +++ exited with 127 +++
>>
>>
>> I can also reproduce this issue on Bullseye (with Linux 5.10.0-21-amd64);
>> while Buster (Linux 4.19.0-23-amd64) is fine.
> [...]
>> ** Command line:
>> root=ZFS=zr/ROOT/debiansid-be ro quiet
>> cgroup_enable=cpuset,cpu,cpuacct,blkio,memory,devices,freezer,net_cls,perf_event,net_prio
>> systemd.unified_cgroup_hierarchy=0
>> net.ifname-policy=keep,onboard,slot,path,kernel zfs.zfs_txg_timeout=60
>> zfs.zfs_arc_max=2166172771 init=/init
> [...]
>
> I can't reproduce this on an ext4 filesystem, so I think ZFS is the
> problem.
>
> ZFS has its own check that blocks a writable mmap of an immutable file,
> without taking MAP_PRIVATE into account:
> https://sources.debian.org/src/zfs-linux/2.1.12-2/module/os/linux/zfs/zfs_vnops_os.c/#L3908
>
> Ben.
>
> --
> Ben Hutchings
> It is easier to change the specification to fit the program
> than vice versa.
>
>
