Re: [PATCH v7 1/4] alloc_tag: move release_module_tags() above reserve_module_tags()

From
sashiko-bot@kernel.org
Date
2026-09-02 08:34:20
Reply to
Hao Ge, 2026-09-02
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] The `release_module_tags()` function modifies the `mod_area_mt` maple tree while holding `mas_lock` (a spinlock), which forces `mas_store()` to use `GFP_NOWAIT` for internal node allocations. If the tree needs to allocate a node (e.g. to merge adjacent `NULL` slots via `mas_wr_extend_null()`) and memory pressure causes the allocation to fail, `mas_store()` silently aborts and leaves the maple state in an error condition. Because `release_module_tags()` does not check `mas_is_err(&mas)` or preallocate nodes, the `mod` pointer remains in the tree after the module is freed. If a subsequently loaded module reuses the exact same `struct module *` address and then is unloaded, `release_module_tags()` will match the stale pointer. This causes it to process the stale tag area, leading to a Use-After-Free when `pr_info` dereferences the old module's freed `.rodata` via `tag->ct.modname`.
--

commit 85214c8b1888956f945e1850b83b26c40f80620f
Author: Hao Ge [off-list ref]

alloc_tag: move release_module_tags() above reserve_module_tags()

This patch moves the release_module_tags() helper function above
reserve_module_tags() in the file. This allows reserve_module_tags() to
call it for cleanup on failure paths without needing a forward declaration.

Cc: <redacted>
quoted hunk ↗ jump to hunk
diff --git a/mm/alloc_tag.c b/mm/alloc_tag.c
index b334103104779..2070e682fe106 100644
--- a/mm/alloc_tag.c
+++ b/mm/alloc_tag.c
[ ... ]
quoted hunk ↗ jump to hunk
@@ -835,6 +835,52 @@ static int vm_module_tags_populate(void)
 	return 0;
 }
 
+static void release_module_tags(struct module *mod, bool used)
+{
+	MA_STATE(mas, &mod_area_mt, module_tags.size, module_tags.size);
[ ... ]
quoted hunk ↗ jump to hunk
+release_area:
+	mas_store(&mas, used ? &unloaded_mod : NULL);
+	val = mas_prev_range(&mas, 0);
+	if (val == &prepend_mod)
+		mas_store(&mas, NULL);
[Severity: High]
This isn't a bug introduced by this patch, but since the code is being
moved here, could these mas_store() calls fail silently under memory
pressure?

Since release_module_tags() modifies the maple tree while holding
mas_lock() (a spinlock), mas_store() uses GFP_NOWAIT for internal node
allocations, such as when it needs to merge adjacent NULL slots.

If a GFP_NOWAIT allocation fails, mas_store() aborts and leaves the
maple state in an error condition. Because this function doesn't
preallocate nodes before acquiring the lock or check mas_is_err(&mas)
after the store, the stale module pointer remains in the tree.

If a subsequently loaded module reuses the exact same struct module
address and is later unloaded, could this cause release_module_tags()
to process the stale entry instead? This appears it would trigger a
Use-After-Free when pr_info() dereferences the old module's freed
.rodata section via tag->ct.modname.
quoted hunk ↗ jump to hunk
+out:
+	mas_unlock(&mas);
+}
-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260902081802.146145-1-hao.ge@linux.dev?part=1
Thread (18 messages) 18 messages, 4 authors, last reply 2026-09-08
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help