Re: [PATCH v2 perf/core] perf script: fix a use after free crash.

[Krister Johansen]

On Tue, Nov 22, 2016 at 04:01:06PM -0300, Arnaldo Carvalho de Melo wrote:
> Sorry for the overly long delay, trying it now after fixing up a
> conflict with a recent patchkit (branch stuff) I tested it by running
> 'perf top -g' and I'm getting some assertion bugs:
> 
> 
> # perf top -g
>            1.34% filemap_map_pages
>          - 0.59% alloc_pages_vma
>               1.20% __alloc_pages_nodemask
> -    5.87%     0.45%  [kernel]                            [k] handle_mm_fault
>    - 1.94% handle_mm_fault
>         1.34% filemap_map_pages
>       - 0.59% alloc_pages_vma
>            1.22% __alloc_pages_nodemask
> +    5.75%     0.03%  perf                                [.] hist_entry_iter__add
> +    4.46%     0.00%  [unknown]                           [.] 0000000000000000
> -    4.06%     2.74%  libc-2.23.so                        [.] _int_malloc
>    - 1.95% 0
>         1.94% _int_malloc
> -    3.20%     0.23%  perf                                [.] iter_add_next_cumulative_entry
>    - 1.49% iter_add_next_cumulative_entry
>       - 1.43% __hists__add_entry
>      2.58%     0.01%  [kernel]                            [k] return_from_SYSCALL_64
>      2.57%     2.55%  libperl.so.5.22.2                   [.] Perl_fbm_instr
> -    2.54%     2.51%  liblzma.so.5.2.2                    [.] lzma_decode
>    - 2.51% lzma_decode
>      2.33%     0.00%  ld-2.23.so                          [.] _dl_sysdep_start
> +    2.24%     0.04%  ld-2.23.so                          [.] dl_main
>      2.13%     0.03%  [kernel]                            [k] ext4_readdir
>      2.09%     0.01%  [kernel]                            [k] sys_newstat
>      2.08%     0.04%  [kernel]                            [k] vfs_fstatat
>      2.07%     0.02%  [kernel]                            [k] SYSC_newstat
>      2.02%     0.01%  [kernel]                            [k] iterate_dir
> -    1.96%     0.17%  [kernel]                            [k] __alloc_pages_nodemask
>    - 1.37% __alloc_pages_nodemask
> perf: util/map.c:246: map__exit: Assertion `!(!((&map->rb_node)->__rb_parent_color == (unsigned long)(&map->rb_node)))' failed.
>                                                                                                                                Aborted (core dumped)
> [root@jouet ~]# 
> 
> 
> I'll try to investigate this further later/tomorrow, find the updated patch below.
> 
> - Arnaldo
> 
> commit af04d2c4a5d1f6bd7f4971118e4e1153cc7c2506
> Author: Krister Johansen <kjlx@xxxxxxxxxxxxxxxxxx>
> Date:   Tue Oct 11 02:28:39 2016 -0700
> 
>     perf callchain: Fix a use after free crash due to refcounting bug
>     
>     If dso__load_kcore frees all of the existing maps, but one has already
>     been attached to a callchain cursor node, then we can get a SIGSEGV in
>     any function that happens to try to use this invalid cursor.  Use the
>     existing map refcount mechanism to forestall cleanup of a map until the
>     cursor iterates past the node.
>     
>     Signed-off-by: Krister Johansen <kjlx@xxxxxxxxxxxxxxxxxx>
>     Cc: Frederic Weisbecker <fweisbec@xxxxxxxxx>
>     Cc: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
>     Cc: Namhyung Kim <namhyung@xxxxxxxxxx>
>     Link: http://lkml.kernel.org/r/20161011092839.GC7837@xxxxxxxxxxxxxxxxxx
>     Signed-off-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
> 
> diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c
> index 823befd8209a..18bb7caee535 100644
> --- a/tools/perf/util/callchain.c
> +++ b/tools/perf/util/callchain.c
> @@ -437,7 +437,7 @@ fill_node(struct callchain_node *node, struct callchain_cursor *cursor)
>  		}
>  		call->ip = cursor_node->ip;
>  		call->ms.sym = cursor_node->sym;
> -		call->ms.map = cursor_node->map;
> +		call->ms.map = map__get(cursor_node->map);
>  
>  		if (cursor_node->branch) {
>  			call->branch_count = 1;
> @@ -477,6 +477,7 @@ add_child(struct callchain_node *parent,
>  
>  		list_for_each_entry_safe(call, tmp, &new->val, list) {
>  			list_del(&call->list);
> +			map__zput(call->ms.map);
>  			free(call);
>  		}
>  		free(new);
> @@ -761,6 +762,7 @@ merge_chain_branch(struct callchain_cursor *cursor,
>  					list->ms.map, list->ms.sym,
>  					false, NULL, 0, 0);
>  		list_del(&list->list);
> +		map__zput(list->ms.map);
>  		free(list);
>  	}
>  
> @@ -811,7 +813,8 @@ int callchain_cursor_append(struct callchain_cursor *cursor,
>  	}
>  
>  	node->ip = ip;
> -	node->map = map;
> +	map__zput(node->map);
> +	node->map = map__get(map);
>  	node->sym = sym;
>  	node->branch = branch;
>  	node->nr_loop_iter = nr_loop_iter;
> @@ -868,6 +871,8 @@ int fill_callchain_info(struct addr_location *al, struct callchain_cursor_node *
>  			goto out;
>  	}
>  
> +	map__get(al->map);
> +
>  	if (al->map->groups == &al->machine->kmaps) {
>  		if (machine__is_host(al->machine)) {
>  			al->cpumode = PERF_RECORD_MISC_KERNEL;
> @@ -1142,11 +1147,13 @@ static void free_callchain_node(struct callchain_node *node)
>  
>  	list_for_each_entry_safe(list, tmp, &node->parent_val, list) {
>  		list_del(&list->list);
> +		map__zput(list->ms.map);
>  		free(list);
>  	}
>  
>  	list_for_each_entry_safe(list, tmp, &node->val, list) {
>  		list_del(&list->list);
> +		map__zput(list->ms.map);
>  		free(list);
>  	}
>  
> @@ -1210,6 +1217,7 @@ int callchain_node__make_parent_list(struct callchain_node *node)
>  				goto out;
>  			*new = *chain;
>  			new->has_children = false;
> +			map__get(new->ms.map);
>  			list_add_tail(&new->list, &head);
>  		}
>  		parent = parent->parent;
> @@ -1230,6 +1238,7 @@ int callchain_node__make_parent_list(struct callchain_node *node)
>  out:
>  	list_for_each_entry_safe(chain, new, &head, list) {
>  		list_del(&chain->list);
> +		map__zput(chain->ms.map);
>  		free(chain);
>  	}
>  	return -ENOMEM;
> diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
> index d9c70dccf06a..f551fd2cfe5a 100644
> --- a/tools/perf/util/callchain.h
> +++ b/tools/perf/util/callchain.h
> @@ -5,6 +5,7 @@
>  #include <linux/list.h>
>  #include <linux/rbtree.h>
>  #include "event.h"
> +#include "map.h"
>  #include "symbol.h"
>  
>  #define HELP_PAD "\t\t\t\t"
> @@ -184,8 +185,13 @@ int callchain_merge(struct callchain_cursor *cursor,
>   */
>  static inline void callchain_cursor_reset(struct callchain_cursor *cursor)
>  {
> +	struct callchain_cursor_node *node;
> +
>  	cursor->nr = 0;
>  	cursor->last = &cursor->first;
> +
> +	for (node = cursor->first; node != NULL; node = node->next)
> +		map__zput(node->map);
>  }
>  
>  int callchain_cursor_append(struct callchain_cursor *cursor, u64 ip,
> diff --git a/tools/perf/util/hist.c b/tools/perf/util/hist.c
> index e1be4132054d..be4b07145705 100644
> --- a/tools/perf/util/hist.c
> +++ b/tools/perf/util/hist.c
> @@ -1,6 +1,7 @@
>  #include "util.h"
>  #include "build-id.h"
>  #include "hist.h"
> +#include "map.h"
>  #include "session.h"
>  #include "sort.h"
>  #include "evlist.h"
> @@ -979,6 +980,7 @@ iter_finish_cumulative_entry(struct hist_entry_iter *iter,
>  {
>  	zfree(&iter->priv);
>  	iter->he = NULL;
> +	map__zput(al->map);
>  
>  	return 0;
>  }


As part of trying to tie up the year-end loose-ends, I went back and
re-tested a rebase'd version of this patch against perf/core.  I ended
up with a merge that's identical to yours, except that I'm not seeing
any assertion failures with 'perf top -g', 'perf script', or 'perf
report'.  Was perf/core the branch that was giving you trouble?

-K