uprobes: use pagesize-aligned virtual address when replacing pages
Subsystems (4)
PATCH 1/1 uprobes: use pagesize-aligned virtual address when replacing pages
uprobes passes an unaligned page mapping address to folio_add_new_anon_rmap(), which ends up triggering a VM_BUG_ON() we recently extended in commit 372cbd4d5a066 ("mm: non-pmd-mappable, large folios for folio_add_new_anon_rmap()").
Arguably, this is uprobes code doing something wrong; however, for the time being it would have likely worked in rmap code because __folio_set_anon() would set folio->index to the same value.
Looking at __replace_page(), we'd also pass slightly wrong values to mmu_notifier_range_init(), page_vma_mapped_walk(), flush_cache_page(), ptep_clear_flush() and set_pte_at_notify(). I suspect most of them are fine, but let's just mark the introducing commit as the one needed fixing. I don't think CC stable is warranted.
We'll add more sanity checks in rmap code separately, to make sure that we always get properly aligned addresses.
--- kernel/events/uprobes.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/events/uprobes.c +1 -1
@@ -537,7 +537,7 @@ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, } } - ret =__replace_page (vma, vaddr, old_page, new_page); + ret =__replace_page (vma, vaddr &PAGE_MASK , old_page, new_page);if (new_page)put_page (new_page); put_old:
Signature · 1 line
2.43.0
kernel/events/uprobes.c int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, · line 540
Quoted 9 lines · David Hildenbrand
kernel/events/uprobes.c +1
-1@@ -537,7 +537,7 @@ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, } } - ret =__replace_page (vma, vaddr, old_page, new_page); + ret =__replace_page (vma, vaddr &PAGE_MASK , old_page, new_page);
cc-ing bpf list
jirka
Quoted 56 lines · David Hildenbrand
uprobes passes an unaligned page mapping address to
folio_add_new_anon_rmap(), which ends up triggering aVM_BUG_ON()we recently extended in commit372cbd4d5a066("mm: non-pmd-mappable, large folios forfolio_add_new_anon_rmap()").Arguably, this is uprobes code doing something wrong; however, for the time being it would have likely worked in rmap code because
__folio_set_anon()would setfolio->indexto the same value.Looking at
__replace_page(), we'd also pass slightly wrong values tommu_notifier_range_init(),page_vma_mapped_walk(),flush_cache_page(),ptep_clear_flush()andset_pte_at_notify(). I suspect most of them are fine, but let's just mark the introducing commit as the one needed fixing. I don't think CC stable is warranted.We'll add more sanity checks in rmap code separately, to make sure that we always get properly aligned addresses.
Reported-by Jiri Olsa <[email protected]>Fixes c517ee744b96 ("uprobes: __replace_page() should not use page_address_in_vma()")Tested-by Jiri Olsa <[email protected]>Reviewed-by Ryan Roberts <[email protected]>Cc Andrew Morton <[email protected]>Cc Oleg Nesterov <[email protected]>Cc Peter Zijlstra <[email protected]>Cc Ingo Molnar <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Mark Rutland <[email protected]>Cc Alexander ShishkinCc Jiri Olsa <[email protected]>Cc Namhyung Kim <[email protected]>Cc Ian Rogers <[email protected]>Cc Adrian Hunter <[email protected]>Signed-off-by David Hildenbrand <[email protected]>--- kernel/events/uprobes.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)kernel/events/uprobes.c +1
-1@@ -537,7 +537,7 @@ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm, } } - ret =__replace_page (vma, vaddr, old_page, new_page); + ret =__replace_page (vma, vaddr &PAGE_MASK , old_page, new_page);if (new_page)put_page (new_page); put_old:Signature · 1 line
2.43.0