Re: [PATCH v2] gdb: rely on the first non-exited thread TPID when reading Linux procfs files

[Simon Marchi <simark@simark.ca>]

On 9/17/26 11:43 AM, Simon Marchi wrote:
> On 8/24/26 12:21 PM, Matthieu Longo wrote:
>> On Linux, /proc/<pid> is keyed by the thread-group leader PID. When
>> the leader has exited, some /proc/<pid>/... entries become unavailable
>> even though another thread is still alive. This can happen, for instance,
>> when the main thread calls pthread_exit() and another thread continues
>> the execution (existing test: gcore-stale-thread).
>>
>> This causes GDB to fail to read procfs entries such as cmdline, cwd,
>> exe, maps, and smaps when it uses 'current_inferior ()->pid' after the
>> thread-group leader has exited.
>>
>> Fix this by adding inferior::first_non_exited_thread(), which returns
>> the PTID of the first non-exited thread of the inferior. Use its LWP ID
> 
> No longer true (it returns the thread_info, not the ptid).
> 
>> when accessing procfs entries that only need a representative live LWP
>> belonging to the process.
>>
>> This is a best-effort choice of a thread that is expected to still exist
>> in the target. Since GDB's view of the threads list may be stale, the
>> selected thread may already have exited by the time it is accessed.
>> Callers must therefore still be prepared to handle that case.
>>
>> Update the following functions:
>>  - linux_info_proc
>>  - linux_process_address_in_memtag_page
>>  - linux_find_memory_regions_full
>>  - linux_fill_prpsinfo
>>  - linux_address_in_shadow_stack_mem_range
>> to use the first non-exited thread's LWP ID instead of the inferior PID
>> when constructing procfs paths.
>>
>> Add a new test in gdb.threads.
>>
>> Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31207
>>
>> Reviewed-by: Thiago Jung Bauermann <thiago.bauermann@linaro.org>
>> ---
>>  gdb/inferior.c                                | 12 +++
>>  gdb/inferior.h                                | 11 +++
>>  gdb/linux-tdep.c                              | 90 +++++++++++++------
>>  ...access-procfs-while-thread-leader-exited.c | 48 ++++++++++
>>  ...cess-procfs-while-thread-leader-exited.exp | 78 ++++++++++++++++
>>  5 files changed, 211 insertions(+), 28 deletions(-)
>>  create mode 100644 gdb/testsuite/gdb.threads/access-procfs-while-thread-leader-exited.c
>>  create mode 100644 gdb/testsuite/gdb.threads/access-procfs-while-thread-leader-exited.exp
>>
>> diff --git a/gdb/inferior.c b/gdb/inferior.c
>> index 8c619ffec5a..f6d77799ac2 100644
>> --- a/gdb/inferior.c
>> +++ b/gdb/inferior.c
>> @@ -250,6 +250,18 @@ inferior::find_thread (ptid_t ptid)
>>  
>>  /* See inferior.h.  */
>>  
>> +thread_info *
>> +inferior::first_non_exited_thread () const
>> +{
>> +  auto it = this->ptid_thread_map.cbegin ();
>> +  if (it != this->ptid_thread_map.cend ())
>> +    return it->second;
>> +  else
>> +    return nullptr;
>> +}
> 
> I was thinking that it would make more sense to name the method
> "any_non_exited_thread", since there's no ordering of the threads
> (ptid_thread_map is an unordered_map).
> 
> Then I saw we already have the free function
> any_non_exited_thread_of_inferior.  It doesn't use the same strategy to
> find a non-exited thread, but the intent seems to be the same.
> 
> But one thing it does is that it prefers the currently selected thread:
> 
>   /* Prefer the current thread, if there's one.  */
>   if (inf == current_inferior () && inferior_ptid != null_ptid)
>     return inferior_thread ();
> 
> I think there is a bug there, because if the selected thread is zombie
> leader (which happens in your test), then it will return it, even if it
> is exited.  We could fix that and then use that function.
> 
> I'm thinking it would actually be nice if GDB prioritized the selected
> thread for /proc access here.  Most things are the same for the whole
> process, which non-exited thread we use is not important for those.  But
> there are some things in /proc that are thread-specific (things like
> stat counters I tihnk), so it would allow the user to select one
> particular thread and then get information about that thread.
> 
> I am trying it, I'll send something soon to the list.

Here it is, could you please give it a look?

https://inbox.sourceware.org/gdb-patches/20260917201755.589524-1-simon.marchi@efficios.com/T/#m8b6bd53bac36790acc05b7475512c6d94b762a53

Thanks,

Simon