bug#51105: 29.0.50; Buffer overflow bug in ns_compute_glyph_string_overhangs

bug#51105: 29.0.50; Buffer overflow bug in ns_compute_glyph_string_overhangs

[Daniel Martín via Bug reports for GNU Emacs, the Swiss army knife of text editors]

There is a buffer overflow bug in the function
ns_compute_glyph_string_overhangs with some particular information
received from the display engine.

(I haven't reduced the test case yet so you may not reproduce the
issue with the following recipe.)

  emacs -Q

Attach a debugger to the Emacs process and add the following
conditional breakpoint:

  br set -f nsterm.m -l 2853 -c 's->nchars==0'

Continue running Emacs

  M-x eww RET wikipedia.org RET

The debugger will stop with the following backtrace:

* thread #1, queue = 'com.apple.main-thread', stop reason = breakpoint 1.1
  * frame #0: 0x000000010e25a20e emacs`ns_compute_glyph_string_overhangs(s=0x00007ffee232ef40) at nsterm.m:2853:7
    frame #1: 0x000000010da4cbdf emacs`draw_glyphs(w=0x00006210000ac130, x=66, row=0x000062b00029ae00, area=TEXT_AREA, start=0, end=12, hl=DRAW_NORMAL_TEXT, overlaps=0) at xdisp.c:29036:4
    frame #2: 0x000000010da49bd0 emacs`gui_write_glyphs(w=0x00006210000ac130, updated_row=0x000062b00029ae00, start=0x0000629001be4200, updated_area=TEXT_AREA, len=12) at xdisp.c:31179:7
    frame #3: 0x000000010d90bc4d emacs`update_text_area(w=0x00006210000ac130, updated_row=0x000062b00029ae00, vpos=28) at dispnew.c:3934:2
    frame #4: 0x000000010d902191 emacs`update_window_line(w=0x00006210000ac130, vpos=28, mouse_face_overwritten_p=0x00007ffee2331720) at dispnew.c:4177:11
    frame #5: 0x000000010d8d84f7 emacs`update_window(w=0x00006210000ac130, force_p=true) at dispnew.c:3680:19
    frame #6: 0x000000010d8d9bbc emacs`update_window_tree(w=0x00006210000ac130, force_p=true) at dispnew.c:3405:14
    frame #7: 0x000000010d8d67e6 emacs`update_frame(f=0x00006210000ad530, force_p=true, inhibit_hairy_id_p=false) at dispnew.c:3240:18
    frame #8: 0x000000010d9db568 emacs`redisplay_internal at xdisp.c:16160:16
    frame #9: 0x000000010d9eb0a9 emacs`redisplay_preserve_echo_area(from_where=12) at xdisp.c:16429:7
    frame #10: 0x000000010e0cb8e1 emacs`wait_reading_process_output(time_limit=0, nsecs=0, read_kbd=-1, do_display=true, wait_for_cell=0x0000000000000000, wait_proc=0x0000000000000000, just_wait_proc=0) at process.c:5789:7
    frame #11: 0x000000010dd99c82 emacs`kbd_buffer_get_event(kbp=0x00007ffee23371c0, used_mouse_menu=0x00007ffee23386c0, end_time=0x0000000000000000) at keyboard.c:3924:4
    frame #12: 0x000000010dd9825e emacs`read_event_from_main_queue(end_time=0x0000000000000000, local_getcjmp=0x00007ffee2338300, used_mouse_menu=0x00007ffee23386c0) at keyboard.c:2198:7
    frame #13: 0x000000010dd6a19a emacs`read_decoded_event_from_main_queue(end_time=0x0000000000000000, local_getcjmp=0x00007ffee2338300, prev_event=0x0000000000000000, used_mouse_menu=0x00007ffee23386c0) at keyboard.c:2262:11
    frame #14: 0x000000010dd632c8 emacs`read_char(commandflag=1, map=0x00006290003eb8a3, prev_event=0x0000000000000000, used_mouse_menu=0x00007ffee23386c0, end_time=0x0000000000000000) at keyboard.c:2892:11
    frame #15: 0x000000010dd58e1d emacs`read_key_sequence(keybuf=0x00007ffee23393a0, prompt=0x0000000000000000, dont_downcase_last=false, can_return_switch_frame=true, fix_current_buffer=true, prevent_redisplay=false) at keyboard.c:9619:12
    frame #16: 0x000000010dd539f3 emacs`command_loop_1 at keyboard.c:1392:15
    frame #17: 0x000000010dfa45d9 emacs`internal_condition_case(bfun=(emacs`command_loop_1 at keyboard.c:1278), handlers=0x0000000000000090, hfun=(emacs`cmd_error at keyboard.c:936)) at eval.c:1453:25
    frame #18: 0x000000010dd52903 emacs`command_loop_2(handlers=0x0000000000000090) at keyboard.c:1133:11
    frame #19: 0x000000010dfa2ff9 emacs`internal_catch(tag=0x000000000000df80, func=(emacs`command_loop_2 at keyboard.c:1129), arg=0x0000000000000090) at eval.c:1184:25
    frame #20: 0x000000010dd50f81 emacs`command_loop at keyboard.c:1111:2
    frame #21: 0x000000010dd50c9b emacs`recursive_edit_1 at keyboard.c:720:9
    frame #22: 0x000000010dd5147a emacs`Frecursive_edit at keyboard.c:803:3
    frame #23: 0x000000010dd4a05a emacs`main(argc=2, argv=0x00007ffee233a310) at emacs.c:2310:3
    frame #24: 0x00007fff20496f3d libdyld.dylib`start + 1

This line in nsterm.m will be executed and is problematic:

  codes[1] = *(s->char2b + s->nchars - 1);

When s->nchars is 0, the code will reference one position before
s->char2b.

I have two questions:

1) Is there any reason the function chooses the first and last glyphs
instead of passing the whole glyph string and rely on text_extents to
perfom boundary checks? That is, I propose:

diff --git a/src/nsterm.m b/src/nsterm.m
index a6c2e7505b..207da60481 100644
--- a/src/nsterm.m
+++ b/src/nsterm.m
@@ -2853,11 +2853,7 @@ Hide the window (X11 semantics)
   if (s->char2b)
     {
       struct font_metrics metrics;
-      unsigned int codes[2];
-      codes[0] = *(s->char2b);
-      codes[1] = *(s->char2b + s->nchars - 1);
-
-      font->driver->text_extents (font, codes, 2, &metrics);
+      font->driver->text_extents (font, s->char2b, s->nchars, &metrics);
       s->left_overhang = -metrics.lbearing;
       s->right_overhang
 	= metrics.rbearing > metrics.width

This way to call the text_extents API is also implemented in w32term.c
and xterm.c.

2) The root cause of the issue may be that s->nchars is 0 when it
shouldn't.  Is there any legitimate scenario where the display engine
may call this routine with s->nchars equal to 0? If so, what are those
situations?


In GNU Emacs 29.0.50 (build 1, x86_64-apple-darwin20.6.0, NS appkit-2022.60 Version 11.6 (Build 20G165))
 of 2021-10-09 built on Daniels-MacBook-Pro.local
Repository revision: 36d7c4af7c83c4f3ea9ab9fdd0822b986564d78e
Repository branch: master
Windowing system distributor 'Apple', version 10.3.2022
System Description:  macOS 11.6

Configured using:
 'configure 'CFLAGS=-O0 -g3''

Configured features:
ACL DBUS GIF GLIB GMP GNUTLS JPEG JSON LCMS2 LIBXML2 MODULES NOTIFY
KQUEUE NS PDUMPER PNG RSVG THREADS TIFF TOOLKIT_SCROLL_BARS XIM ZLIB

Important settings:
  value of $LANG: en_US.UTF-8
  locale-coding-system: utf-8-unix

Major mode: Lisp Interaction

Minor modes in effect:
  tooltip-mode: t
  global-eldoc-mode: t
  eldoc-mode: t
  show-paren-mode: t
  electric-indent-mode: t
  mouse-wheel-mode: t
  tool-bar-mode: t
  menu-bar-mode: t
  file-name-shadow-mode: t
  global-font-lock-mode: t
  font-lock-mode: t
  blink-cursor-mode: t
  auto-composition-mode: t
  auto-encryption-mode: t
  auto-compression-mode: t
  line-number-mode: t
  indent-tabs-mode: t
  transient-mark-mode: t

Load-path shadows:
None found.

Features:
(shadow sort mail-extr emacsbug message rmc puny dired dired-loaddefs
rfc822 mml mml-sec epa derived epg rfc6068 epg-config gnus-util rmail
rmail-loaddefs auth-source cl-seq eieio eieio-core cl-macs
eieio-loaddefs password-cache json map text-property-search time-date
seq gv subr-x byte-opt bytecomp byte-compile cconv mm-decode mm-bodies
mm-encode mail-parse rfc2231 mailabbrev gmm-utils mailheader cl-loaddefs
cl-lib sendmail rfc2047 rfc2045 ietf-drums mm-util mail-prsvr mail-utils
iso-transl tooltip eldoc paren electric uniquify ediff-hook vc-hooks
lisp-float-type elisp-mode mwheel term/ns-win ns-win ucs-normalize
mule-util term/common-win tool-bar dnd fontset image regexp-opt fringe
tabulated-list replace newcomment text-mode lisp-mode prog-mode register
page tab-bar menu-bar rfn-eshadow isearch easymenu timer select
scroll-bar mouse jit-lock font-lock syntax font-core term/tty-colors
frame minibuffer cl-generic cham georgian utf-8-lang misc-lang
vietnamese tibetan thai tai-viet lao korean japanese eucjp-ms cp51932
hebrew greek romanian slovak czech european ethiopic indian cyrillic
chinese composite emoji-zwj charscript charprop case-table epa-hook
jka-cmpr-hook help simple abbrev obarray cl-preloaded nadvice button
loaddefs faces cus-face macroexp files window text-properties overlay
sha1 md5 base64 format env code-pages mule custom widget
hashtable-print-readable backquote threads dbusbind kqueue cocoa ns
lcms2 multi-tty make-network-process emacs)

Memory information:
((conses 16 49678 8809)
 (symbols 48 6572 1)
 (strings 32 17870 1691)
 (string-bytes 1 591830)
 (vectors 16 12905)
 (vector-slots 8 177066 9811)
 (floats 8 21 51)
 (intervals 56 191 0)
 (buffers 992 10))

bug#51105: 29.0.50; Buffer overflow bug in ns_compute_glyph_string_overhangs

[Eli Zaretskii]

> Date: Sat, 09 Oct 2021 02:30:33 +0200
> From:  Daniel Martín via "Bug reports for GNU Emacs,
>  the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>
> 
> 2) The root cause of the issue may be that s->nchars is 0 when it
> shouldn't.  Is there any legitimate scenario where the display engine
> may call this routine with s->nchars equal to 0? If so, what are those
> situations?

I think if the glyph string has composition glyphs, nchars can be
zero.  What is the value of s->first_glyph->type in the case where it
happens?

bug#51105: 29.0.50; Buffer overflow bug in ns_compute_glyph_string_overhangs

[Daniel Martín via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Eli Zaretskii <eliz@gnu.org> writes:

>> Date: Sat, 09 Oct 2021 02:30:33 +0200
>> From:  Daniel Martín via "Bug reports for GNU Emacs,
>>  the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>
>> 
>> 2) The root cause of the issue may be that s->nchars is 0 when it
>> shouldn't.  Is there any legitimate scenario where the display engine
>> may call this routine with s->nchars equal to 0? If so, what are those
>> situations?
>
> I think if the glyph string has composition glyphs, nchars can be
> zero.  What is the value of s->first_glyph->type in the case where it
> happens?

Yep, it seems so:

(lldb) fr v s->first_glyph->type
(unsigned int:3) s->first_glyph->type = 1

I've found a 2006 commit that seemed to handle this particular pointer
arithmetic logic for when the type of the first glyph is STRETCH_GLYPH:
https://git.savannah.gnu.org/cgit/emacs.git/commit/?id=825de9a1027073beaec38ab1572e9d954f8a1eb0

Now I think that the right thing to do may be to modify nsterm.m, switch
on the glyph type and, if the glyph type is COMPOSITE_GLYPH, call
composition_gstring_width to get the glyph metrics.  Function
composition_gstring_width uses the values from fields s->cmp_from and
s->cmp_to, and would avoid the buffer overflow:

(lldb) fr v s->cmp_from
(int) s->cmp_from = 6
(lldb) fr v s->cmp_to
(int) s->cmp_to = 7

WDYT? I can prepare a patch of this type if you agree.

I'll try to get the sequence of codepoints from the glyph string in the
debugger, so we can have a reduced test case (ie. the exact string from
the Wikipedia's front page that causes the issue).

bug#51105: 29.0.50; Buffer overflow bug in ns_compute_glyph_string_overhangs

[Eli Zaretskii]

> From: Daniel Martín <mardani29@yahoo.es>
> Cc: 51105@debbugs.gnu.org
> Date: Sat, 09 Oct 2021 12:06:36 +0200
> 
> Now I think that the right thing to do may be to modify nsterm.m, switch
> on the glyph type and, if the glyph type is COMPOSITE_GLYPH, call
> composition_gstring_width to get the glyph metrics.  Function
> composition_gstring_width uses the values from fields s->cmp_from and
> s->cmp_to, and would avoid the buffer overflow:
> 
> (lldb) fr v s->cmp_from
> (int) s->cmp_from = 6
> (lldb) fr v s->cmp_to
> (int) s->cmp_to = 7
> 
> WDYT? I can prepare a patch of this type if you agree.

SGTM, but I'd like to hear Alan's opinion as well, as I don't feel I
know enough about the NS display backend.

bug#51105: 29.0.50; Buffer overflow bug in ns_compute_glyph_string_overhangs

[Alan Third]

On Sat, Oct 09, 2021 at 02:43:18PM +0300, Eli Zaretskii wrote:
> > From: Daniel Martín <mardani29@yahoo.es>
> > Cc: 51105@debbugs.gnu.org
> > Date: Sat, 09 Oct 2021 12:06:36 +0200
> > 
> > Now I think that the right thing to do may be to modify nsterm.m, switch
> > on the glyph type and, if the glyph type is COMPOSITE_GLYPH, call
> > composition_gstring_width to get the glyph metrics.  Function
> > composition_gstring_width uses the values from fields s->cmp_from and
> > s->cmp_to, and would avoid the buffer overflow:
> > 
> > (lldb) fr v s->cmp_from
> > (int) s->cmp_from = 6
> > (lldb) fr v s->cmp_to
> > (int) s->cmp_to = 7
> > 
> > WDYT? I can prepare a patch of this type if you agree.
> 
> SGTM, but I'd like to hear Alan's opinion as well, as I don't feel I
> know enough about the NS display backend.

I don't know much about this part of the code, but it sounds good to
me too.
-- 
Alan Third

bug#51105: 29.0.50; Buffer overflow bug in ns_compute_glyph_string_overhangs

[Daniel Martín via Bug reports for GNU Emacs, the Swiss army knife of text editors]

[-- Attachment #1: Type: text/plain, Size: 1135 bytes --]

Alan Third <alan@idiocy.org> writes:

> On Sat, Oct 09, 2021 at 02:43:18PM +0300, Eli Zaretskii wrote:
>> > From: Daniel Martín <mardani29@yahoo.es>
>> > Cc: 51105@debbugs.gnu.org
>> > Date: Sat, 09 Oct 2021 12:06:36 +0200
>> > 
>> > Now I think that the right thing to do may be to modify nsterm.m, switch
>> > on the glyph type and, if the glyph type is COMPOSITE_GLYPH, call
>> > composition_gstring_width to get the glyph metrics.  Function
>> > composition_gstring_width uses the values from fields s->cmp_from and
>> > s->cmp_to, and would avoid the buffer overflow:
>> > 
>> > (lldb) fr v s->cmp_from
>> > (int) s->cmp_from = 6
>> > (lldb) fr v s->cmp_to
>> > (int) s->cmp_to = 7
>> > 
>> > WDYT? I can prepare a patch of this type if you agree.
>> 
>> SGTM, but I'd like to hear Alan's opinion as well, as I don't feel I
>> know enough about the NS display backend.
>
> I don't know much about this part of the code, but it sounds good to
> me too.

A reduced test case to reproduce the problem is to paste "العربية" in the
*scratch* buffer.

I've attached a patch that fixes the issue.


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-Fix-buffer-overflow-in-ns_compute_glyph_string_overh.patch --]
[-- Type: text/x-patch, Size: 2437 bytes --]

From 23897a25d7ddebc06ab855058d36a5e291e5cba3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Daniel=20Mart=C3=ADn?= <mardani29@yahoo.es>
Date: Sat, 9 Oct 2021 21:10:20 +0200
Subject: [PATCH] Fix buffer overflow in ns_compute_glyph_string_overhangs

* src/nsterm.m (ns_compute_glyph_string_overhangs): When the first
glyph of a glyph string is a composite glyph, `s->nchars' is 0, so
"s->char2b + s->nchars - 1" dereferenced a position before buffer
`s->char2b'.  Instead, rewrite part of the function to distinguish
between character glyphs and composite glyphs.  For character glyphs,
calculate the font metrics using the `text_extents' function, passing
it the entire glyph string; for composite glyphs, call
`composition_gstring_width'. (Bug#51105)
---
 src/nsterm.m | 29 ++++++++++++++++++-----------
 1 file changed, 18 insertions(+), 11 deletions(-)

diff --git a/src/nsterm.m b/src/nsterm.m
index a6c2e7505b..e616766ec7 100644
--- a/src/nsterm.m
+++ b/src/nsterm.m
@@ -2848,20 +2848,27 @@ Hide the window (X11 semantics)
      External (RIF); compute left/right overhang of whole string and set in s
    -------------------------------------------------------------------------- */
 {
-  struct font *font = s->font;
-
   if (s->char2b)
     {
       struct font_metrics metrics;
-      unsigned int codes[2];
-      codes[0] = *(s->char2b);
-      codes[1] = *(s->char2b + s->nchars - 1);
-
-      font->driver->text_extents (font, codes, 2, &metrics);
-      s->left_overhang = -metrics.lbearing;
-      s->right_overhang
-	= metrics.rbearing > metrics.width
-	? metrics.rbearing - metrics.width : 0;
+      if (s->first_glyph->type == CHAR_GLYPH && !s->font_not_found_p)
+        {
+          struct font *font = s->font;
+          font->driver->text_extents (font, s->char2b, s->nchars, &metrics);
+          s->left_overhang = -metrics.lbearing;
+          s->right_overhang
+            = metrics.rbearing > metrics.width
+            ? metrics.rbearing - metrics.width : 0;
+        }
+      else if (s->first_glyph->type == COMPOSITE_GLYPH)
+        {
+          Lisp_Object gstring = composition_gstring_from_id (s->cmp_id);
+
+	  composition_gstring_width (gstring, s->cmp_from, s->cmp_to, &metrics);
+	  s->right_overhang = (metrics.rbearing > metrics.width
+			       ? metrics.rbearing - metrics.width : 0);
+	  s->left_overhang = metrics.lbearing < 0 ? -metrics.lbearing : 0;
+        }
     }
   else
     {
-- 
2.31.0


[-- Attachment #3: Type: text/plain, Size: 79 bytes --]


Let me know if you like it and please install it on my behalf if so.
Thanks.

bug#51105: 29.0.50; Buffer overflow bug in ns_compute_glyph_string_overhangs

[Daniel Martín via Bug reports for GNU Emacs, the Swiss army knife of text editors]

[-- Attachment #1: Type: text/plain, Size: 464 bytes --]

Daniel Martín via "Bug reports for GNU Emacs, the Swiss army knife of
text editors" <bug-gnu-emacs@gnu.org> writes:

>
> A reduced test case to reproduce the problem is to paste "العربية" in the
> *scratch* buffer.
>
> I've attached a patch that fixes the issue.
>
>
>
> Let me know if you like it and please install it on my behalf if so.
> Thanks.

Sorry, there was an indentation problem in the previous patch.  Here's
an updated one.


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-Fix-buffer-overflow-in-ns_compute_glyph_string_overh.patch --]
[-- Type: text/x-patch, Size: 2479 bytes --]

From 1f64cf0bb78b77570d60f70c2e2342c2293a5ffb Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Daniel=20Mart=C3=ADn?= <mardani29@yahoo.es>
Date: Sat, 9 Oct 2021 21:10:20 +0200
Subject: [PATCH] Fix buffer overflow in ns_compute_glyph_string_overhangs

* src/nsterm.m (ns_compute_glyph_string_overhangs): When the first
glyph of a glyph string is a composite glyph, `s->nchars' is 0, so
"s->char2b + s->nchars - 1" dereferenced a position before buffer
`s->char2b'.  Instead, rewrite part of the function to distinguish
between character glyphs and composite glyphs.  For character glyphs,
calculate the font metrics using the `text_extents' function, passing
it the entire glyph string; for composite glyphs, call
`composition_gstring_width'. (Bug#51105)
---
 src/nsterm.m | 29 ++++++++++++++++++-----------
 1 file changed, 18 insertions(+), 11 deletions(-)

diff --git a/src/nsterm.m b/src/nsterm.m
index a6c2e7505b..e8e08640c6 100644
--- a/src/nsterm.m
+++ b/src/nsterm.m
@@ -2848,20 +2848,27 @@ Hide the window (X11 semantics)
      External (RIF); compute left/right overhang of whole string and set in s
    -------------------------------------------------------------------------- */
 {
-  struct font *font = s->font;
-
   if (s->char2b)
     {
       struct font_metrics metrics;
-      unsigned int codes[2];
-      codes[0] = *(s->char2b);
-      codes[1] = *(s->char2b + s->nchars - 1);
-
-      font->driver->text_extents (font, codes, 2, &metrics);
-      s->left_overhang = -metrics.lbearing;
-      s->right_overhang
-	= metrics.rbearing > metrics.width
-	? metrics.rbearing - metrics.width : 0;
+      if (s->first_glyph->type == CHAR_GLYPH && !s->font_not_found_p)
+        {
+          struct font *font = s->font;
+          font->driver->text_extents (font, s->char2b, s->nchars, &metrics);
+          s->left_overhang = -metrics.lbearing;
+          s->right_overhang
+            = metrics.rbearing > metrics.width
+            ? metrics.rbearing - metrics.width : 0;
+        }
+      else if (s->first_glyph->type == COMPOSITE_GLYPH)
+        {
+          Lisp_Object gstring = composition_gstring_from_id (s->cmp_id);
+
+          composition_gstring_width (gstring, s->cmp_from, s->cmp_to, &metrics);
+          s->right_overhang = (metrics.rbearing > metrics.width
+                               ? metrics.rbearing - metrics.width : 0);
+          s->left_overhang = metrics.lbearing < 0 ? -metrics.lbearing : 0;
+        }
     }
   else
     {
-- 
2.31.0

bug#51105: 29.0.50; Buffer overflow bug in ns_compute_glyph_string_overhangs

[Lars Ingebrigtsen]

Daniel Martín <mardani29@yahoo.es> writes:

> Sorry, there was an indentation problem in the previous patch.  Here's
> an updated one.

It seemed like Alan agreed with the fix, and I tested it now on my M1
Apple laptop, and it didn't break anything obvious, so I've now pushed
Daniel's patch to the trunk.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no