binfmt_elf: Don't write past end of notes for regset gap
authorRick Edgecombe <rick.p.edgecombe@intel.com>
Thu, 17 Mar 2022 19:20:13 +0000 (12:20 -0700)
committerKees Cook <keescook@chromium.org>
Fri, 18 Mar 2022 17:17:09 +0000 (10:17 -0700)
In fill_thread_core_info() the ptrace accessible registers are collected
to be written out as notes in a core file. The note array is allocated
from a size calculated by iterating the user regset view, and counting the
regsets that have a non-zero core_note_type. However, this only allows for
there to be non-zero core_note_type at the end of the regset view. If
there are any gaps in the middle, fill_thread_core_info() will overflow the
note allocation, as it iterates over the size of the view and the
allocation would be smaller than that.

There doesn't appear to be any arch that has gaps such that they exceed
the notes allocation, but the code is brittle and tries to support
something it doesn't. It could be fixed by increasing the allocation size,
but instead just have the note collecting code utilize the array better.
This way the allocation can stay smaller.

Even in the case of no arch's that have gaps in their regset views, this
introduces a change in the resulting indicies of t->notes. It does not
introduce any changes to the core file itself, because any blank notes are
skipped in write_note_info().

In case, the allocation logic between fill_note_info() and
fill_thread_core_info() ever diverges from the usage logic, warn and skip
writing any notes that would overflow the array.

This fix is derrived from an earlier one[0] by Yu-cheng Yu.

[0] https://lore.kernel.org/lkml/20180717162502.32274-1-yu-cheng.yu@intel.com/

Co-developed-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20220317192013.13655-4-rick.p.edgecombe@intel.com
fs/binfmt_elf.c

index ca5296cae9791af714ba93ac339774a1b6708514..37d9c455d53510d2f4a87d2245f01292e0d1b0d8 100644 (file)
@@ -1766,9 +1766,9 @@ static void do_thread_regset_writeback(struct task_struct *task,
 
 static int fill_thread_core_info(struct elf_thread_core_info *t,
                                 const struct user_regset_view *view,
-                                long signr, size_t *total)
+                                long signr, struct elf_note_info *info)
 {
-       unsigned int i;
+       unsigned int note_iter, view_iter;
 
        /*
         * NT_PRSTATUS is the one special case, because the regset data
@@ -1782,17 +1782,17 @@ static int fill_thread_core_info(struct elf_thread_core_info *t,
 
        fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
                  PRSTATUS_SIZE, &t->prstatus);
-       *total += notesize(&t->notes[0]);
+       info->size += notesize(&t->notes[0]);
 
        do_thread_regset_writeback(t->task, &view->regsets[0]);
 
        /*
         * Each other regset might generate a note too.  For each regset
-        * that has no core_note_type or is inactive, we leave t->notes[i]
-        * all zero and we'll know to skip writing it later.
+        * that has no core_note_type or is inactive, skip it.
         */
-       for (i = 1; i < view->n; ++i) {
-               const struct user_regset *regset = &view->regsets[i];
+       note_iter = 1;
+       for (view_iter = 1; view_iter < view->n; ++view_iter) {
+               const struct user_regset *regset = &view->regsets[view_iter];
                int note_type = regset->core_note_type;
                bool is_fpreg = note_type == NT_PRFPREG;
                void *data;
@@ -1808,13 +1808,17 @@ static int fill_thread_core_info(struct elf_thread_core_info *t,
                if (ret < 0)
                        continue;
 
+               if (WARN_ON_ONCE(note_iter >= info->thread_notes))
+                       break;
+
                if (is_fpreg)
                        SET_PR_FPVALID(&t->prstatus);
 
-               fill_note(&t->notes[i], is_fpreg ? "CORE" : "LINUX",
+               fill_note(&t->notes[note_iter], is_fpreg ? "CORE" : "LINUX",
                          note_type, ret, data);
 
-               *total += notesize(&t->notes[i]);
+               info->size += notesize(&t->notes[note_iter]);
+               note_iter++;
        }
 
        return 1;
@@ -1894,7 +1898,7 @@ static int fill_note_info(struct elfhdr *elf, int phdrs,
         * Now fill in each thread's information.
         */
        for (t = info->thread; t != NULL; t = t->next)
-               if (!fill_thread_core_info(t, view, cprm->siginfo->si_signo, &info->size))
+               if (!fill_thread_core_info(t, view, cprm->siginfo->si_signo, info))
                        return 0;
 
        /*