elf: Always call destructors in reverse constructor order (bug 30785)

The current implementation of dlclose (and process exit) re-sorts the
link maps before calling ELF destructors.  Destructor order is not the
reverse of the constructor order as a result: The second sort takes
relocation dependencies into account, and other differences can result
from ambiguous inputs, such as cycles.  (The force_first handling in
_dl_sort_maps is not effective for dlclose.)  After the changes in
this commit, there is still a required difference due to
dlopen/dlclose ordering by the application, but the previous
discrepancies went beyond that.

A new global (namespace-spanning) list of link maps,
_dl_init_called_list, is updated right before ELF constructors are
called from _dl_init.

In dl_close_worker, the maps variable, an on-stack variable length
array, is eliminated.  (VLAs are problematic, and dlclose should not
call malloc because it cannot readily deal with malloc failure.)
Marking still-used objects uses the namespace list directly, with
next and next_idx replacing the done_index variable.

After marking, _dl_init_called_list is used to call the destructors
of now-unused maps in reverse destructor order.  These destructors
can call dlopen.  Previously, new objects do not have l_map_used set.
This had to change: There is no copy of the link map list anymore,
so processing would cover newly opened (and unmarked) mappings,
unloading them.  Now, _dl_init (indirectly) sets l_map_used, too.
(dlclose is handled by the existing reentrancy guard.)

After _dl_init_called_list traversal, two more loops follow.  The
processing order changes to the original link map order in the
namespace.  Previously, dependency order was used.  The difference
should not matter because relocation dependencies could already
reorder link maps in the old code.

The changes to _dl_fini remove the sorting step and replace it with
a traversal of _dl_init_called_list.  The l_direct_opencount
decrement outside the loader lock is removed because it appears
incorrect: the counter manipulation could race with other dynamic
loader operations.

tst-audit23 needs adjustments to the changes in LA_ACT_DELETE
notifications.  The new approach for checking la_activity should
make it clearer that la_activty calls come in pairs around namespace
updates.

The dependency sorting test cases need updates because the destructor
order is always the opposite order of constructor order, even with
relocation dependencies or cycles present.

There is a future cleanup opportunity to remove the now-constant
force_first and for_fini arguments from the _dl_sort_maps function.

Fixes commit 1df71d32fe5f5905ffd5d100e5e9ca8ad62 ("elf: Implement
force_first handling in _dl_sort_maps_dfs (bug 28937)").

Reviewed-by: DJ Delorie <dj@redhat.com>

1 files changed, 6 insertions, 13 deletions

diff --git a/elf/dso-sort-tests-1.def b/elf/dso-sort-tests-1.def
index 4bf9052db1..61dc54f8ae 100644
--- a/elf/dso-sort-tests-1.def
+++ b/elf/dso-sort-tests-1.def
@@ -53,21 +53,14 @@ tst-dso-ordering10: {}->a->b->c;soname({})=c
 output: b>a>{}<a<b
 
 # Complex example from Bugzilla #15311, under-linked and with circular
-# relocation(dynamic) dependencies. While this is technically unspecified, the
-# presumed reasonable practical behavior is for the destructor order to respect
-# the static DT_NEEDED links (here this means the a->b->c->d order).
-# The older dynamic_sort=1 algorithm does not achieve this, while the DFS-based
-# dynamic_sort=2 algorithm does, although it is still arguable whether going
-# beyond spec to do this is the right thing to do.
-# The below expected outputs are what the two algorithms currently produce
-# respectively, for regression testing purposes.
+# relocation(dynamic) dependencies. For both sorting algorithms, the
+# destruction order is the reverse of the construction order, and
+# relocation dependencies are not taken into account.
 tst-bz15311: {+a;+e;+f;+g;+d;%d;-d;-g;-f;-e;-a};a->b->c->d;d=>[ba];c=>a;b=>e=>a;c=>f=>b;d=>g=>c
-output(glibc.rtld.dynamic_sort=1): {+a[d>c>b>a>];+e[e>];+f[f>];+g[g>];+d[];%d(b(e(a()))a()g(c(a()f(b(e(a()))))));-d[];-g[];-f[];-e[];-a[<a<c<d<g<f<b<e];}
-output(glibc.rtld.dynamic_sort=2): {+a[d>c>b>a>];+e[e>];+f[f>];+g[g>];+d[];%d(b(e(a()))a()g(c(a()f(b(e(a()))))));-d[];-g[];-f[];-e[];-a[<g<f<a<b<c<d<e];}
+output: {+a[d>c>b>a>];+e[e>];+f[f>];+g[g>];+d[];%d(b(e(a()))a()g(c(a()f(b(e(a()))))));-d[];-g[];-f[];-e[];-a[<g<f<e<a<b<c<d];}
 
 # Test that even in the presence of dependency loops involving dlopen'ed
 # object, that object is initialized last (and not unloaded prematurely).
-# Final destructor order is indeterminate due to the cycle.
+# Final destructor order is the opposite of constructor order.
 tst-bz28937: {+a;+b;-b;+c;%c};a->a1;a->a2;a2->a;b->b1;c->a1;c=>a1
-output(glibc.rtld.dynamic_sort=1): {+a[a2>a1>a>];+b[b1>b>];-b[<b<b1];+c[c>];%c(a1());}<a<a2<c<a1
-output(glibc.rtld.dynamic_sort=2): {+a[a2>a1>a>];+b[b1>b>];-b[<b<b1];+c[c>];%c(a1());}<a2<a<c<a1
+output: {+a[a2>a1>a>];+b[b1>b>];-b[<b<b1];+c[c>];%c(a1());}<c<a<a1<a2