inspect io_uring_peek_batch_cqe cqes memory in gdb

[Comrade88]

i'm trying to explore the contents of cqe in the debugger while debugging the application. gdb says it can't read the memory.
investigation showed that this memory is mmap'ed on the io_uring file descriptor, but ptrace access is not implemented.
is it planned to implement filling vma->vm_ops in io_uring_mmap with something that can provide ptrace access to the returned
memory, maybe there is some other way to examine its contents from the debugger in user space?

[isilence]

There is nothing too special about the backing memory, print cqe[0] works just fine. Are there any clues why it can't read it?

[Comrade88]

print cqe[0] returns a pointer to an io_uring_cqe struct and this pointer from the mmap'ed region is not resolved in either gdb or lldb.

[rohnet]

example of error on attempt to resolve pointer to cqe in lldb:

error: Couldn't apply expression side effects : Couldn't dematerialize a result variable: couldn't read its memory

[isilence]

(gdb) p cqe
$1 = (struct io_uring_cqe *) 0x7ffff7fc0040
(gdb) p sqe
$2 = (struct io_uring_sqe *) 0x7ffff7fbf000
(gdb) p cqe[0]
$3 = {user_data = 1, res = 0, flags = 0, big_cqe = 0x7ffff7fc0050}
(gdb) p sqe[0]
$4 = {opcode = 0 '\000', flags = 0 '\000', ioprio = 0, fd = -1, {off = 0, addr2 = 0, {cmd_op = 0, __pad1 = 0}}, {addr = 0, splice_off_in = 0, {level = 0, optname = 0}}, len = 0, {rw_flags = 0, fsync_flags = 0, poll_events = 0, 
    poll32_events = 0, sync_range_flags = 0, msg_flags = 0, timeout_flags = 0, accept_flags = 0, cancel_flags = 0, open_flags = 0, statx_flags = 0, fadvise_advice = 0, splice_flags = 0, rename_flags = 0, unlink_flags = 0, 
    hardlink_flags = 0, xattr_flags = 0, msg_ring_flags = 0, uring_cmd_flags = 0, waitid_flags = 0, futex_flags = 0, install_fd_flags = 0, nop_flags = 0}, user_data = 1, {buf_index = 0, buf_group = 0}, personality = 0, {
    splice_fd_in = 0, file_index = 0, optlen = 0, {addr_len = 0, __pad3 = {0}}}, {{addr3 = 0, __pad2 = {0}}, optval = 0, cmd = 0x7ffff7fbf030 ""}}

Works just fine for me. Can you print the pointers and while it's still stopped do cat /proc/<pid>/maps for me? Is it x86?

[Comrade88]

ok. yes, it is x86. which kernel are u using? we expirience the problem on 5.15

[rohnet]

2 peeked CQEs:

(gdb) p nr
$3 = 2
(gdb) p buf[0]
$4 = (io_uring_cqe *) 0x7fffefade140
(gdb) p buf[1]
$5 = (io_uring_cqe *) 0x7fffefade150
(gdb) p *buf[0]
Cannot access memory at address 0x7fffefade140
(gdb) p *buf[1]
Cannot access memory at address 0x7fffefade150

from proc maps:

7fffefada000-7fffefade000 rw-s 10000000 00:0e 12612                      anon_inode:[io_uring]
7fffefade000-7fffefae1000 rw-s 00000000 00:0e 12612                      anon_inode:[io_uring]

[YoSTEALTH]

I have had io_uring_peek_batch_cqe segfault for many years now even in Linux 6.13, could this issue be related? https://github.com/YoSTEALTH/Liburing/blob/master/test/queue/get_cqe_test.py#L25-L42 (comment out @pytest.mark.skip('BUG') and run test_peek_batch).