From: Lancelot SIX <lancelot.six@amd.com>
To: Simon Marchi <simon.marchi@efficios.com>, <gdb-patches@sourceware.org>
Subject: Re: [PATCH v2 5/10] gdb/solib-rocm: move per-inferior data to rocm_solib_ops
Date: Mon, 6 Jul 2026 18:09:30 +0100 [thread overview]
Message-ID: <6fq2jwcy7nc6ande7vbiuxkd3n2u2nrrvzvezw2ivok6epyewc@zqfl6jj7qneb> (raw)
In-Reply-To: <20260608200100.666134-6-simon.marchi@efficios.com>
On Mon, Jun 08, 2026 at 04:00:29PM -0400, Simon Marchi wrote:
> Move the data currently stored in an inferior registry directly into
> rocm_solib_ops. This changes the storage of this data from "per
> inferior" to "per program-space", but it should be equivalent, since
> there is usually a 1:1 mapping between those two. The only time there
> isn't a 1:1 mapping is during a vfork, before the exec/exit, but no ROCm
> activity happens during that slice of time.
>
> Function rocm_update_solib_list becomes
> rocm_solib_ops::update_solib_list.
>
> Function rocm_bfd_iovec_open becomes rocm_solib_ops::bfd_iovec_open.
>
> Most of the changes consist of accessing the fields directly, instead of
> going through the solib_info object.
>
> The rocm_solib_ops constructor has to take an inferior as a parameter,
> rather than a program_space, because of the fd cache.
>
Hi Simon,
It seems that this patch introduces a use-after-free issue when handling
the "detach-on-fork on", "follow-fork-mode child" case (I see this with
the gdb.rocm/fork-exec-gpu-to-non-gpu.exp testcase with ASAN enabled).
When doing the case where we follow the child, we reach
rocm_solib_target_inferior_forked:
if (detach_on_fork && follow_child && fork_kind == TARGET_WAITKIND_FORKED)
{
auto rocm_ops_holder = child_inf->pspace->release_solib_ops ();
auto rocm_ops
= gdb::checked_static_cast<rocm_solib_ops *> (rocm_ops_holder.get ());
child_inf->pspace->set_solib_ops (rocm_ops->release_host_ops ());
}
On exit here of this block, the rocm_ops is deleted (calls
rocm_solib_ops::~rocm_solib_ops).
However, at this stage, if there are still solibs which were opened by
rocm_solib_ops, the next call to update_solib_list will lead to deleting
the solib which do not exist anymore (so far so good). Any file
rocm-solib still open at this point will eventually be closed when we
close the associated BFD. This calls into rocm_solib_fd_cache::close,
which tries to access the rocm_solib_fd_cache::m_cache field. However,
this is a reference to a rocm_solib_ops::m_fd_cache, which is now gone,
leading to a use after free.
This was not an issue until this patch because the fd_cache was owned by
the per-inferior rocm_solib_data. which is not going anywhere during the fork.
Best,
Lancelot.
> Change-Id: I386139008b6f0eca1d897fc35be18faa9d5a104e
> ---
> gdb/solib-rocm.c | 90 ++++++++++++++++++------------------------------
> 1 file changed, 33 insertions(+), 57 deletions(-)
>
> diff --git a/gdb/solib-rocm.c b/gdb/solib-rocm.c
> index 4032ada28a0f..618f4f5e7905 100644
> --- a/gdb/solib-rocm.c
> +++ b/gdb/solib-rocm.c
> @@ -181,30 +181,15 @@ struct rocm_so
> lm_info_svr4_up lm_info;
> };
>
> -struct solib_info
> -{
> - explicit solib_info (inferior *inf)
> - : fd_cache (inf)
> - {};
> -
> - /* List of code objects loaded into the inferior. */
> - std::vector<rocm_so> solib_list;
> -
> - /* Cache of opened FD in the inferior. */
> - rocm_solib_fd_cache fd_cache;
> -};
> -
> -/* Per-inferior data key. */
> -static const registry<inferior>::key<solib_info> rocm_solib_data;
> -
> /* solib_ops for ROCm systems. */
>
> struct rocm_solib_ops : public solib_ops
> {
> /* HOST_OPS is the host solib_ops that rocm_solib_ops hijacks / wraps,
> in order to provide support for ROCm code objects. */
> - explicit rocm_solib_ops (program_space *pspace, solib_ops_up host_ops)
> - : solib_ops (pspace), m_host_ops (std::move (host_ops))
> + explicit rocm_solib_ops (inferior *inf, solib_ops_up host_ops)
> + : solib_ops (inf->pspace), m_host_ops (std::move (host_ops)),
> + m_fd_cache (inf)
> {
> gdb_assert (m_host_ops != nullptr);
> gdb_assert (dynamic_cast<rocm_solib_ops *> (m_host_ops.get ()) == nullptr);
> @@ -221,6 +206,10 @@ struct rocm_solib_ops : public solib_ops
> void relocate_section_addresses (solib &so, target_section *) const override;
> void handle_event () override;
>
> + /* This needs to be public because it is called from
> + rocm_solib_target_inferior_created. */
> + void update_solib_list ();
> +
> /* Implement the following methods just to forward the calls to the host
> solib_ops. We currently need to implement all the methods that
> svr4_solib_ops implements. */
> @@ -271,18 +260,17 @@ struct rocm_solib_ops : public solib_ops
> private:
> owning_intrusive_list<solib>
> solibs_from_rocm_sos (const std::vector<rocm_so> &sos);
> + gdb_bfd_iovec_base *bfd_iovec_open (bfd *abfd, inferior *inferior);
>
> solib_ops_up m_host_ops;
> +
> + /* List of code objects loaded into the inferior. */
> + std::vector<rocm_so> m_solib_list;
> +
> + /* Cache of opened FD in the inferior. */
> + rocm_solib_fd_cache m_fd_cache;
> };
>
> -/* Fetch the solib_info data for INF. */
> -
> -static struct solib_info *
> -get_solib_info (inferior *inf)
> -{
> - return &rocm_solib_data.try_emplace (inf, inf);
> -}
> -
> /* Relocate section addresses. */
>
> void
> @@ -300,8 +288,6 @@ rocm_solib_ops::relocate_section_addresses (solib &so,
> sec->endaddr = sec->endaddr + li->l_addr;
> }
>
> -static void rocm_update_solib_list ();
> -
> void
> rocm_solib_ops::handle_event ()
> {
> @@ -313,7 +299,7 @@ rocm_solib_ops::handle_event ()
> previously loaded). */
> m_host_ops->handle_event ();
>
> - rocm_update_solib_list ();
> + this->update_solib_list ();
> }
>
> /* Create solib objects from rocm_so objects in SOS. */
> @@ -340,12 +326,10 @@ rocm_solib_ops::current_sos ()
> owning_intrusive_list<solib> sos = m_host_ops->current_sos ();
>
> /* Then, the device-side shared library list. */
> - std::vector<rocm_so> &dev_sos = get_solib_info (current_inferior ())->solib_list;
> -
> - if (dev_sos.empty ())
> + if (m_solib_list.empty ())
> return sos;
>
> - owning_intrusive_list<solib> dev_solibs = solibs_from_rocm_sos (dev_sos);
> + owning_intrusive_list<solib> dev_solibs = solibs_from_rocm_sos (m_solib_list);
>
> if (sos.empty ())
> return dev_solibs;
> @@ -525,8 +509,8 @@ rocm_code_object_stream_memory::read (bfd *, void *buf, file_ptr size,
>
> } /* anonymous namespace */
>
> -static gdb_bfd_iovec_base *
> -rocm_bfd_iovec_open (bfd *abfd, inferior *inferior)
> +gdb_bfd_iovec_base *
> +rocm_solib_ops::bfd_iovec_open (bfd *abfd, inferior *inferior)
> {
> std::string_view uri (bfd_get_filename (abfd));
> std::string_view protocol_delim = "://";
> @@ -635,10 +619,9 @@ rocm_bfd_iovec_open (bfd *abfd, inferior *inferior)
> && decoded_path[3] == '/')
> decoded_path.erase (0, 1);
>
> - auto info = get_solib_info (inferior);
> fileio_error target_errno;
> rocm_solib_fd_cache::cached_target_fd fd
> - = info->fd_cache.open (decoded_path, &target_errno);
> + = m_fd_cache.open (decoded_path, &target_errno);
>
> if (fd.fd () == target_fd::INVALID)
> {
> @@ -698,9 +681,9 @@ rocm_solib_ops::bfd_open (const char *pathname)
> if (strstr (pathname, "://") == nullptr)
> return m_host_ops->bfd_open (pathname);
>
> - auto open = [] (bfd *nbfd) -> gdb_bfd_iovec_base *
> + auto open = [this] (bfd *nbfd)
> {
> - return rocm_bfd_iovec_open (nbfd, current_inferior ());
> + return this->bfd_iovec_open (nbfd, current_inferior ());
> };
>
> gdb_bfd_ref_ptr abfd = gdb_bfd_openr_iovec (pathname, "elf64-amdgcn", open);
> @@ -782,13 +765,12 @@ rocm_solib_ops::bfd_open (const char *pathname)
> void
> rocm_solib_ops::create_inferior_hook (int from_tty)
> {
> - get_solib_info (current_inferior ())->solib_list.clear ();
> -
> + m_solib_list.clear ();
> m_host_ops->create_inferior_hook (from_tty);
> }
>
> -static void
> -rocm_update_solib_list ()
> +void
> +rocm_solib_ops::update_solib_list ()
> {
> inferior *inf = current_inferior ();
>
> @@ -796,10 +778,7 @@ rocm_update_solib_list ()
> if (process_id.handle == AMD_DBGAPI_PROCESS_NONE.handle)
> return;
>
> - solib_info *info = get_solib_info (inf);
> -
> - info->solib_list.clear ();
> - std::vector<rocm_so> &sos = info->solib_list;
> + m_solib_list.clear ();
>
> amd_dbgapi_code_object_id_t *code_object_list;
> size_t count;
> @@ -846,7 +825,8 @@ rocm_update_solib_list ()
> std::string unique_name
> = string_printf ("code_object_%ld", code_object_list[i].handle);
>
> - sos.emplace_back (uri_bytes, std::move (unique_name), std::move (li));
> + m_solib_list.emplace_back (uri_bytes, std::move (unique_name),
> + std::move (li));
> }
> }
>
> @@ -857,17 +837,15 @@ rocm_solib_target_inferior_created (inferior *inf)
> if (inf->vfork_parent != nullptr)
> return;
>
> - get_solib_info (inf)->solib_list.clear ();
> -
> auto prev_ops = inf->pspace->release_solib_ops ();
> - auto rocm_ops
> - = std::make_unique<rocm_solib_ops> (inf->pspace, std::move (prev_ops));
> + auto rocm_ops = std::make_unique<rocm_solib_ops> (inf, std::move (prev_ops));
> inf->pspace->set_solib_ops (std::move (rocm_ops));
>
> - rocm_update_solib_list ();
> + gdb::checked_static_cast<rocm_solib_ops *>
> + (inf->pspace->solib_ops ())->update_solib_list ();
>
> /* Force GDB to reload the solibs. */
> - current_inferior ()->pspace->clear_solib_cache ();
> + inf->pspace->clear_solib_cache ();
> solib_add (nullptr, 0, auto_solib_add);
> }
>
> @@ -882,10 +860,8 @@ rocm_solib_target_inferior_execd (inferior *exec_inf, inferior *follow_inf)
> auto pspace = follow_inf->pspace;
> auto prev_ops = pspace->release_solib_ops ();
> auto rocm_ops
> - = std::make_unique<rocm_solib_ops> (pspace, std::move (prev_ops));
> + = std::make_unique<rocm_solib_ops> (follow_inf, std::move (prev_ops));
> pspace->set_solib_ops (std::move (rocm_ops));
> -
> - get_solib_info (exec_inf)->solib_list.clear ();
> }
>
> static void
> --
> 2.54.0
next prev parent reply other threads:[~2026-07-06 17:10 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-06-08 20:00 [PATCH RESEND v2 00/10] Multiple solib_ops in a program_space Simon Marchi
2026-06-08 20:00 ` [PATCH RESEND v2 01/10] gdb/solib-rocm: assert that host ops isn't rocm_solib_ops Simon Marchi
2026-07-06 17:08 ` [PATCH v2 1/10] " Lancelot SIX
2026-06-08 20:00 ` [PATCH RESEND v2 02/10] gdb/solib-rocm: pass reference to cache to rocm_code_object_stream_file Simon Marchi
2026-06-08 20:00 ` [PATCH RESEND v2 03/10] gdb/solib-rocm: add cached_target_fd to manage cached fd lifetime Simon Marchi
2026-07-06 17:08 ` [PATCH v2 3/10] " Lancelot SIX
2026-07-07 20:48 ` Simon Marchi
2026-06-08 20:00 ` [PATCH RESEND v2 04/10] gdb: de-constify some methods of solib_ops Simon Marchi
2026-07-06 17:09 ` [PATCH v2 4/10] " Lancelot SIX
2026-06-08 20:00 ` [PATCH RESEND v2 05/10] gdb/solib-rocm: move per-inferior data to rocm_solib_ops Simon Marchi
2026-07-06 17:09 ` Lancelot SIX [this message]
2026-07-08 16:08 ` [PATCH v2 5/10] " Simon Marchi
2026-06-08 20:00 ` [PATCH RESEND v2 06/10] gdb/solib-rocm: save inferior in rocm_solib_ops Simon Marchi
2026-07-06 17:10 ` [PATCH v2 6/10] " Lancelot SIX
2026-06-08 20:00 ` [PATCH RESEND v2 07/10] gdb/solib: add remove_solib function Simon Marchi
2026-06-08 20:00 ` [PATCH RESEND v2 08/10] gdb: add objfile -> solib backlink Simon Marchi
2026-06-08 20:00 ` [PATCH RESEND v2 09/10] gdb: change default objfile iteration order to start with current objfile Simon Marchi
2026-06-08 20:00 ` [PATCH RESEND v2 10/10] gdb: multiple solib_ops per program space Simon Marchi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=6fq2jwcy7nc6ande7vbiuxkd3n2u2nrrvzvezw2ivok6epyewc@zqfl6jj7qneb \
--to=lancelot.six@amd.com \
--cc=gdb-patches@sourceware.org \
--cc=simon.marchi@efficios.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox