CVE-2023-53668

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> ring-buffer: Fix deadloop issue on reading trace_pipe<br /> <br /> Soft lockup occurs when reading file &amp;#39;trace_pipe&amp;#39;:<br /> <br /> watchdog: BUG: soft lockup - CPU#6 stuck for 22s! [cat:4488]<br /> [...]<br /> RIP: 0010:ring_buffer_empty_cpu+0xed/0x170<br /> RSP: 0018:ffff88810dd6fc48 EFLAGS: 00000246<br /> RAX: 0000000000000000 RBX: 0000000000000246 RCX: ffffffff93d1aaeb<br /> RDX: ffff88810a280040 RSI: 0000000000000008 RDI: ffff88811164b218<br /> RBP: ffff88811164b218 R08: 0000000000000000 R09: ffff88815156600f<br /> R10: ffffed102a2acc01 R11: 0000000000000001 R12: 0000000051651901<br /> R13: 0000000000000000 R14: ffff888115e49500 R15: 0000000000000000<br /> [...]<br /> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033<br /> CR2: 00007f8d853c2000 CR3: 000000010dcd8000 CR4: 00000000000006e0<br /> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000<br /> DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400<br /> Call Trace:<br /> __find_next_entry+0x1a8/0x4b0<br /> ? peek_next_entry+0x250/0x250<br /> ? down_write+0xa5/0x120<br /> ? down_write_killable+0x130/0x130<br /> trace_find_next_entry_inc+0x3b/0x1d0<br /> tracing_read_pipe+0x423/0xae0<br /> ? tracing_splice_read_pipe+0xcb0/0xcb0<br /> vfs_read+0x16b/0x490<br /> ksys_read+0x105/0x210<br /> ? __ia32_sys_pwrite64+0x200/0x200<br /> ? switch_fpu_return+0x108/0x220<br /> do_syscall_64+0x33/0x40<br /> entry_SYSCALL_64_after_hwframe+0x61/0xc6<br /> <br /> Through the vmcore, I found it&amp;#39;s because in tracing_read_pipe(),<br /> ring_buffer_empty_cpu() found some buffer is not empty but then it<br /> cannot read anything due to "rb_num_of_entries() == 0" always true,<br /> Then it infinitely loop the procedure due to user buffer not been<br /> filled, see following code path:<br /> <br /> tracing_read_pipe() {<br /> ... ...<br /> waitagain:<br /> tracing_wait_pipe() // 1. find non-empty buffer here<br /> trace_find_next_entry_inc() // 2. loop here try to find an entry<br /> __find_next_entry()<br /> ring_buffer_empty_cpu(); // 3. find non-empty buffer<br /> peek_next_entry() // 4. but peek always return NULL<br /> ring_buffer_peek()<br /> rb_buffer_peek()<br /> rb_get_reader_page()<br /> // 5. because rb_num_of_entries() == 0 always true here<br /> // then return NULL<br /> // 6. user buffer not been filled so goto &amp;#39;waitgain&amp;#39;<br /> // and eventually leads to an deadloop in kernel!!!<br /> }<br /> <br /> By some analyzing, I found that when resetting ringbuffer, the &amp;#39;entries&amp;#39;<br /> of its pages are not all cleared (see rb_reset_cpu()). Then when reducing<br /> the ringbuffer, and if some reduced pages exist dirty &amp;#39;entries&amp;#39; data, they<br /> will be added into &amp;#39;cpu_buffer-&gt;overrun&amp;#39; (see rb_remove_pages()), which<br /> cause wrong &amp;#39;overrun&amp;#39; count and eventually cause the deadloop issue.<br /> <br /> To fix it, we need to clear every pages in rb_reset_cpu().