public inbox for gdb@sourceware.org
 help / color / mirror / Atom feed
* [PATCH] GDB: Fix the overflow in addr_is_displayed()
@ 2020-01-04 11:43 Shahab Vahedi
  2020-01-06  0:30 ` Andrew Burgess
  0 siblings, 1 reply; 2+ messages in thread
From: Shahab Vahedi @ 2020-01-04 11:43 UTC (permalink / raw)
  To: gdb-patches; +Cc: gdb, Shahab Vahedi, Claudiu Zissulescu, Francois Bedard

From: Shahab Vahedi <shahab@synopsys.com>

In a corner case scenario, where the height of the assembly TUI is
bigger than the number of instructions in the whole program, GDB
dumps core. The problem roots in this condition check:

  int i = 0;
  while (i < content. size() - threshold ...) {
    ... content[i] ...
  }

"threshold" is 2 and there are times that "content. size()" is 0.
This results into an overflow and the loop is entered whereas it
should have been skipped.

This has been discussed at length in bug 25345:
  https://sourceware.org/bugzilla/show_bug.cgi?id=25345

gdb/ChangeLog:
2020-01-04  Shahab Vahedi  <shahab@synopsys.com>
        * tui/tui-disasm.c (tui_disasm_window::addr_is_displayed):
        Treat "content.size()" as "int" to avoid overflow.
---
 gdb/tui/tui-disasm.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/gdb/tui/tui-disasm.c b/gdb/tui/tui-disasm.c
index c72b50730b0..a0921eb09d1 100644
--- a/gdb/tui/tui-disasm.c
+++ b/gdb/tui/tui-disasm.c
@@ -349,10 +349,10 @@ bool
 tui_disasm_window::addr_is_displayed (CORE_ADDR addr) const
 {
   bool is_displayed = false;
-  int threshold = SCROLL_THRESHOLD;
+  int nr_of_lines = int(content.size()) - int(SCROLL_THRESHOLD);
 
   int i = 0;
-  while (i < content.size () - threshold && !is_displayed)
+  while (i < nr_of_lines && !is_displayed)
     {
       is_displayed
 	= (content[i].line_or_addr.loa == LOA_ADDRESS
-- 
2.24.1

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH] GDB: Fix the overflow in addr_is_displayed()
  2020-01-04 11:43 [PATCH] GDB: Fix the overflow in addr_is_displayed() Shahab Vahedi
@ 2020-01-06  0:30 ` Andrew Burgess
  0 siblings, 0 replies; 2+ messages in thread
From: Andrew Burgess @ 2020-01-06  0:30 UTC (permalink / raw)
  To: Shahab Vahedi
  Cc: gdb-patches, gdb, Shahab Vahedi, Claudiu Zissulescu, Francois Bedard

* Shahab Vahedi <shahab.vahedi@gmail.com> [2020-01-04 12:43:12 +0100]:

> From: Shahab Vahedi <shahab@synopsys.com>
> 
> In a corner case scenario, where the height of the assembly TUI is
> bigger than the number of instructions in the whole program, GDB
> dumps core. The problem roots in this condition check:
> 
>   int i = 0;
>   while (i < content. size() - threshold ...) {
>     ... content[i] ...
>   }
> 
> "threshold" is 2 and there are times that "content. size()" is 0.
> This results into an overflow and the loop is entered whereas it
> should have been skipped.

I didn't quite understand the problem description.

I can see how 'content.size() - threshold' would overflow when the
size is 0 or 1, but I don't understand the part about the tui height
being bigger than the number of instructions.

I tried to reproduce the failure on native x86-64 but failed.  I can
get the initial stage reproduced, where the asm window is empty and I
see the error about "Cannot access memory at address ....", but then
when I attach to a remote the asm window fills in fine.  I guess this
is because on Linux the page the code is on is readable, and under
QEMU only the specific program instructions are readable maybe?

Anyway, the change looks reasonable, though I had one comment, inline
below...

> 
> This has been discussed at length in bug 25345:
>   https://sourceware.org/bugzilla/show_bug.cgi?id=25345
> 
> gdb/ChangeLog:
> 2020-01-04  Shahab Vahedi  <shahab@synopsys.com>
>         * tui/tui-disasm.c (tui_disasm_window::addr_is_displayed):
>         Treat "content.size()" as "int" to avoid overflow.
> ---
>  gdb/tui/tui-disasm.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/gdb/tui/tui-disasm.c b/gdb/tui/tui-disasm.c
> index c72b50730b0..a0921eb09d1 100644
> --- a/gdb/tui/tui-disasm.c
> +++ b/gdb/tui/tui-disasm.c
> @@ -349,10 +349,10 @@ bool
>  tui_disasm_window::addr_is_displayed (CORE_ADDR addr) const
>  {
>    bool is_displayed = false;
> -  int threshold = SCROLL_THRESHOLD;
> +  int nr_of_lines = int(content.size()) - int(SCROLL_THRESHOLD);

I had a look through our code and I couldn't find any examples of us
using 'int (xxx)' syntax to cast to int, this is probably a result of
our C heritage, but we should probably try to be consistent I think.
I'd suggest maybe this become:

  int nr_of_lines = (int) content.size () - SCROLL_THRESHOLD;

Thanks,
Andrew

>  
>    int i = 0;
> -  while (i < content.size () - threshold && !is_displayed)
> +  while (i < nr_of_lines && !is_displayed)
>      {
>        is_displayed
>  	= (content[i].line_or_addr.loa == LOA_ADDRESS
> -- 
> 2.24.1
> 

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2020-01-06  0:30 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-01-04 11:43 [PATCH] GDB: Fix the overflow in addr_is_displayed() Shahab Vahedi
2020-01-06  0:30 ` Andrew Burgess

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).