[PATCH 1/2] [gdb/tui] Fix segfault in tui_find_disassembly_address

[PATCH 1/2] [gdb/tui] Fix segfault in tui_find_disassembly_address

[Tom de Vries]

PR29040 describes a FAIL for test-case gdb.threads/next-fork-other-thread.exp
and target board unix/-m32.

The FAIL happens due to the test executable running into an assert, which is
caused by a forked child segfaulting, like so:
...
 Program terminated with signal SIGSEGV, Segmentation fault.
 #0  0x00000000 in ?? ()
...

I tried to reproduce the segfault with exec next-fork-other-thread-fork, using
TUI layout asm.

I set a breakpoint at fork and ran to the breakpoint, and somewhere during the
following session I ran into a gdb segfault here in
tui_find_disassembly_address:
...
	  /* Disassemble forward.  */
	  next_addr = tui_disassemble (gdbarch, asm_lines, new_low, max_lines);
	  last_addr = asm_lines.back ().addr;
...
due to asm_lines being empty after the call to tui_disassemble, while
asm_lines.back () assumes that it's not empty.

I have not been able to reproduce that segfault in that original setting, I'm
not sure of the exact scenario (though looking back it probably involved
"set detach-on-fork off").

What likely happened is that I managed to reproduce PR29040, and TUI (attempted
to) display the disassembly for address 0, which led to the gdb segfault.

When gdb_print_insn encounters an insn it cannot print because it can't read
the memory, it throws a MEMORY_ERROR that is caught by tui_disassemble.

The specific bit that causes the gdb segfault is that if gdb_print_insn throws
a MEMORY_ERROR for the first insn in tui_disassemble, it returns an empty
asm_lines.

FWIW, I did manage to reproduce the gdb segfault as follows:
...
$ gdb -q \
    -iex "set pagination off" \
    /usr/bin/rustc \
    -ex "set breakpoint pending on" \
    -ex "b dl_main" \
    -ex run \
    -ex "up 4" \
    -ex "layout asm" \
    -ex "print \$pc"
  ...
<TUI>
  ...
$1 = (void (*)()) 0x1
(gdb)
...
Now press <up>, and the segfault triggers.

Fix the segfault by handling asm_lines.empty () results of tui_disassemble in
tui_find_disassembly_address.

I've written a unit test that exercises this scenario.

Tested on x86_64-linux.

PR tui/30823
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30823
---
 gdb/tui/tui-disasm.c | 39 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 39 insertions(+)

diff --git a/gdb/tui/tui-disasm.c b/gdb/tui/tui-disasm.c
index f0b55769d71..03c78aa1291 100644
--- a/gdb/tui/tui-disasm.c
+++ b/gdb/tui/tui-disasm.c
@@ -41,6 +41,8 @@
 #include "objfiles.h"
 #include "cli/cli-style.h"
 #include "tui/tui-location.h"
+#include "gdbsupport/selftest.h"
+#include "inferior.h"
 
 #include "gdb_curses.h"
 
@@ -203,6 +205,8 @@ tui_find_disassembly_address (struct gdbarch *gdbarch, CORE_ADDR pc, int from)
 	 instruction fails to disassemble we will take the address of the
 	 previous instruction that did disassemble as the result.  */
       tui_disassemble (gdbarch, asm_lines, pc, max_lines + 1);
+      if (asm_lines.empty ())
+	return pc;
       new_low = asm_lines.back ().addr;
     }
   else
@@ -244,6 +248,8 @@ tui_find_disassembly_address (struct gdbarch *gdbarch, CORE_ADDR pc, int from)
 
 	  /* Disassemble forward.  */
 	  next_addr = tui_disassemble (gdbarch, asm_lines, new_low, max_lines);
+	  if (asm_lines.empty ())
+	    break;
 	  last_addr = asm_lines.back ().addr;
 
 	  /* If disassembling from the current value of NEW_LOW reached PC
@@ -522,3 +528,36 @@ tui_disasm_window::display_start_addr (struct gdbarch **gdbarch_p,
   *gdbarch_p = m_gdbarch;
   *addr_p = m_start_line_or_addr.u.addr;
 }
+
+#if GDB_SELF_TEST
+namespace selftests {
+namespace tui {
+namespace disasm {
+
+static void
+run_tests ()
+{
+  if (current_inferior () != nullptr)
+    {
+      struct gdbarch *gdbarch = current_inferior ()->gdbarch;
+
+      /* Check that tui_find_disassembly_address robustly handles the case of
+	 being passed a PC for which gdb_print_insn throws a MEMORY_ERROR.  */
+      SELF_CHECK (tui_find_disassembly_address (gdbarch, 0, 1) == 0);
+      SELF_CHECK (tui_find_disassembly_address (gdbarch, 0, -1) == 0);
+    }
+}
+
+} /* namespace disasm */
+} /* namespace tui */
+} /* namespace selftests */
+#endif /* GDB_SELF_TEST */
+
+void _initialize_tui_disasm ();
+void
+_initialize_tui_disasm ()
+{
+#if GDB_SELF_TEST
+  selftests::register_test ("tui-disasm", selftests::tui::disasm::run_tests);
+#endif
+}

base-commit: b6ac461ace19ba19aaf135a028df4e67e47e21d7
-- 
2.35.3

[PATCH 2/2] [gdb/tui] Only handle code sections in tui_find_backward_disassembly_start_address

[Tom de Vries]

After adding a unit test in gdb/tui/tui-disasm.c excercising
tui_find_disassembly_address, I decided to try to extend it using addresses
around section borders.

The new test was very slow (when using gdb as inferior, as is done in
gdb.gdb/unittest.exp), due to disassembling entire non-code sections.

Fix this this by limiting tui_find_backward_disassembly_start_address to
SEC_CODE sections.

FWIW, compared to other self-tests it's still somewhat slow:
...
(gdb) maint selftest
  ...
Command execution time: 1.535391 (cpu), 1.571246 (wall)
(gdb) maint selftest tui-disasm
  ...
Command execution time: 0.482022 (cpu), 0.482028 (wall)
...
This is for calling gdb_print_insn ~550 times.

Tested on x86_64-linux.
---
 gdb/tui/tui-disasm.c | 50 +++++++++++++++++++++++++++++++++++++-------
 1 file changed, 43 insertions(+), 7 deletions(-)

diff --git a/gdb/tui/tui-disasm.c b/gdb/tui/tui-disasm.c
index 03c78aa1291..c31ab5b0680 100644
--- a/gdb/tui/tui-disasm.c
+++ b/gdb/tui/tui-disasm.c
@@ -46,6 +46,8 @@
 
 #include "gdb_curses.h"
 
+#include <unordered_set>
+
 struct tui_asm_line
 {
   CORE_ADDR addr;
@@ -164,6 +166,15 @@ tui_disassemble (struct gdbarch *gdbarch,
 static CORE_ADDR
 tui_find_backward_disassembly_start_address (CORE_ADDR addr)
 {
+  struct obj_section *section = find_pc_section (addr);
+  /* Don't handle addresses not in a known section.  */
+  if (section == nullptr)
+    return addr;
+
+  /* Only handle sections with only code.  */
+  if ((section->the_bfd_section->flags & SEC_CODE) == 0)
+    return addr;
+
   struct bound_minimal_symbol msym, msym_prev;
 
   msym = lookup_minimal_symbol_by_pc_section (addr - 1, nullptr,
@@ -174,13 +185,8 @@ tui_find_backward_disassembly_start_address (CORE_ADDR addr)
   else if (msym_prev.minsym != nullptr)
     return msym_prev.value_address ();
 
-  /* Find the section that ADDR is in, and look for the start of the
-     section.  */
-  struct obj_section *section = find_pc_section (addr);
-  if (section != NULL)
-    return section->addr ();
-
-  return addr;
+  /* Use the start of the section.  */
+  return section->addr ();
 }
 
 /* Find the disassembly address that corresponds to FROM lines above
@@ -545,6 +551,36 @@ run_tests ()
 	 being passed a PC for which gdb_print_insn throws a MEMORY_ERROR.  */
       SELF_CHECK (tui_find_disassembly_address (gdbarch, 0, 1) == 0);
       SELF_CHECK (tui_find_disassembly_address (gdbarch, 0, -1) == 0);
+
+      /* Poke around the edges of sections.  */
+      gdbarch_iterate_over_objfiles_in_search_order
+	(target_gdbarch (),
+	 [gdbarch] (objfile *obj)
+	 {
+	   std::unordered_set<CORE_ADDR> visited;
+
+	   /* Already done above.  */
+	   visited.insert (0);
+
+	   for (obj_section *osect : obj->sections ())
+	     {
+	       CORE_ADDR first_addr = osect->addr ();
+	       CORE_ADDR last_addr = osect->endaddr () - 1;
+
+	       for (auto addr_ : { first_addr, last_addr })
+		 for (int offset = -1; offset <= 1; ++offset)
+		   {
+		     CORE_ADDR addr = addr_ + offset;
+		     if (visited.find (addr) != visited.end ())
+		       continue;
+
+		     tui_find_disassembly_address (gdbarch, addr, 1);
+		     tui_find_disassembly_address (gdbarch, addr, -1);
+		     visited.insert (addr);
+		   }
+	     }
+	   return false;
+	 }, nullptr);
     }
 }
 
-- 
2.35.3

[PING][PATCH 1/2] [gdb/tui] Fix segfault in tui_find_disassembly_address

[Tom de Vries]

On 9/5/23 17:03, Tom de Vries via Gdb-patches wrote:
> PR29040 describes a FAIL for test-case gdb.threads/next-fork-other-thread.exp
> and target board unix/-m32.
> 
> The FAIL happens due to the test executable running into an assert, which is
> caused by a forked child segfaulting, like so:
> ...
>   Program terminated with signal SIGSEGV, Segmentation fault.
>   #0  0x00000000 in ?? ()
> ...
> 
> I tried to reproduce the segfault with exec next-fork-other-thread-fork, using
> TUI layout asm.
> 
> I set a breakpoint at fork and ran to the breakpoint, and somewhere during the
> following session I ran into a gdb segfault here in
> tui_find_disassembly_address:
> ...
> 	  /* Disassemble forward.  */
> 	  next_addr = tui_disassemble (gdbarch, asm_lines, new_low, max_lines);
> 	  last_addr = asm_lines.back ().addr;
> ...
> due to asm_lines being empty after the call to tui_disassemble, while
> asm_lines.back () assumes that it's not empty.
> 
> I have not been able to reproduce that segfault in that original setting, I'm
> not sure of the exact scenario (though looking back it probably involved
> "set detach-on-fork off").
> 
> What likely happened is that I managed to reproduce PR29040, and TUI (attempted
> to) display the disassembly for address 0, which led to the gdb segfault.
> 
> When gdb_print_insn encounters an insn it cannot print because it can't read
> the memory, it throws a MEMORY_ERROR that is caught by tui_disassemble.
> 
> The specific bit that causes the gdb segfault is that if gdb_print_insn throws
> a MEMORY_ERROR for the first insn in tui_disassemble, it returns an empty
> asm_lines.
> 
> FWIW, I did manage to reproduce the gdb segfault as follows:
> ...
> $ gdb -q \
>      -iex "set pagination off" \
>      /usr/bin/rustc \
>      -ex "set breakpoint pending on" \
>      -ex "b dl_main" \
>      -ex run \
>      -ex "up 4" \
>      -ex "layout asm" \
>      -ex "print \$pc"
>    ...
> <TUI>
>    ...
> $1 = (void (*)()) 0x1
> (gdb)
> ...
> Now press <up>, and the segfault triggers.
> 
> Fix the segfault by handling asm_lines.empty () results of tui_disassemble in
> tui_find_disassembly_address.
> 
> I've written a unit test that exercises this scenario.
> 
> Tested on x86_64-linux.
> 

Ping for both patches in the series.

Thanks,
- Tom

> PR tui/30823
> Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30823
> ---
>   gdb/tui/tui-disasm.c | 39 +++++++++++++++++++++++++++++++++++++++
>   1 file changed, 39 insertions(+)
> 
> diff --git a/gdb/tui/tui-disasm.c b/gdb/tui/tui-disasm.c
> index f0b55769d71..03c78aa1291 100644
> --- a/gdb/tui/tui-disasm.c
> +++ b/gdb/tui/tui-disasm.c
> @@ -41,6 +41,8 @@
>   #include "objfiles.h"
>   #include "cli/cli-style.h"
>   #include "tui/tui-location.h"
> +#include "gdbsupport/selftest.h"
> +#include "inferior.h"
>   
>   #include "gdb_curses.h"
>   
> @@ -203,6 +205,8 @@ tui_find_disassembly_address (struct gdbarch *gdbarch, CORE_ADDR pc, int from)
>   	 instruction fails to disassemble we will take the address of the
>   	 previous instruction that did disassemble as the result.  */
>         tui_disassemble (gdbarch, asm_lines, pc, max_lines + 1);
> +      if (asm_lines.empty ())
> +	return pc;
>         new_low = asm_lines.back ().addr;
>       }
>     else
> @@ -244,6 +248,8 @@ tui_find_disassembly_address (struct gdbarch *gdbarch, CORE_ADDR pc, int from)
>   
>   	  /* Disassemble forward.  */
>   	  next_addr = tui_disassemble (gdbarch, asm_lines, new_low, max_lines);
> +	  if (asm_lines.empty ())
> +	    break;
>   	  last_addr = asm_lines.back ().addr;
>   
>   	  /* If disassembling from the current value of NEW_LOW reached PC
> @@ -522,3 +528,36 @@ tui_disasm_window::display_start_addr (struct gdbarch **gdbarch_p,
>     *gdbarch_p = m_gdbarch;
>     *addr_p = m_start_line_or_addr.u.addr;
>   }
> +
> +#if GDB_SELF_TEST
> +namespace selftests {
> +namespace tui {
> +namespace disasm {
> +
> +static void
> +run_tests ()
> +{
> +  if (current_inferior () != nullptr)
> +    {
> +      struct gdbarch *gdbarch = current_inferior ()->gdbarch;
> +
> +      /* Check that tui_find_disassembly_address robustly handles the case of
> +	 being passed a PC for which gdb_print_insn throws a MEMORY_ERROR.  */
> +      SELF_CHECK (tui_find_disassembly_address (gdbarch, 0, 1) == 0);
> +      SELF_CHECK (tui_find_disassembly_address (gdbarch, 0, -1) == 0);
> +    }
> +}
> +
> +} /* namespace disasm */
> +} /* namespace tui */
> +} /* namespace selftests */
> +#endif /* GDB_SELF_TEST */
> +
> +void _initialize_tui_disasm ();
> +void
> +_initialize_tui_disasm ()
> +{
> +#if GDB_SELF_TEST
> +  selftests::register_test ("tui-disasm", selftests::tui::disasm::run_tests);
> +#endif
> +}
> 
> base-commit: b6ac461ace19ba19aaf135a028df4e67e47e21d7

Re: [PATCH 1/2] [gdb/tui] Fix segfault in tui_find_disassembly_address

[Kevin Buettner]

Hi Tom,

On Tue,  5 Sep 2023 17:03:38 +0200
Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> wrote:

> Fix the segfault by handling asm_lines.empty () results of tui_disassemble in
> tui_find_disassembly_address.
> 
> I've written a unit test that exercises this scenario.
> 
> Tested on x86_64-linux.
> 
> PR tui/30823
> Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30823

I don't know this area very well, but your explanation and patch look
reasonable to me.

Reviewed-by: Kevin Buettner <kevinb@redhat.com>

Re: [PATCH 2/2] [gdb/tui] Only handle code sections in tui_find_backward_disassembly_start_address

[Kevin Buettner]

On Tue,  5 Sep 2023 17:03:39 +0200
Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> wrote:

> Fix this this by limiting tui_find_backward_disassembly_start_address to
> SEC_CODE sections.

What will happen when a user attempts to use the TUI to look at / debug
dynamically generated code?

Kevin

Re: [PATCH 2/2] [gdb/tui] Only handle code sections in tui_find_backward_disassembly_start_address

[Tom de Vries]

On 9/27/23 18:15, Kevin Buettner wrote:
> On Tue,  5 Sep 2023 17:03:39 +0200
> Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> wrote:
> 
>> Fix this this by limiting tui_find_backward_disassembly_start_address to
>> SEC_CODE sections.
> 
> What will happen when a user attempts to use the TUI to look at / debug
> dynamically generated code?
> 

Hi, and thanks for the review(s).

Hm, I don't know, good question.

Maybe it will be possible to detect these cases and expand the search, 
and by default do a more narrow search.

I'm not familiar with the dynamically generated code scenarios and 
support in gdb, so I'm dropping this for now.

Thanks,
- Tom

Re: [PATCH 1/2] [gdb/tui] Fix segfault in tui_find_disassembly_address

[Tom de Vries]

On 9/27/23 18:10, Kevin Buettner wrote:
> Hi Tom,
> 
> On Tue,  5 Sep 2023 17:03:38 +0200
> Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> wrote:
> 
>> Fix the segfault by handling asm_lines.empty () results of tui_disassemble in
>> tui_find_disassembly_address.
>>
>> I've written a unit test that exercises this scenario.
>>
>> Tested on x86_64-linux.
>>
>> PR tui/30823
>> Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30823
> 
> I don't know this area very well, but your explanation and patch look
> reasonable to me.
> 
> Reviewed-by: Kevin Buettner <kevinb@redhat.com>

Thanks for the review.

I've committed this, but I now see there's a new warning:
...
gdb/tui/tui-disasm.c: In function ‘CORE_ADDR 
tui_find_disassembly_address(gdbarch*, CORE_ADDR, int)’:
gdb/tui/tui-disasm.c:293:7: warning: ‘last_addr’ may be used 
uninitialized in this function [-Wmaybe-uninitialized]
        if (last_addr < pc)
        ^~
...

I'll take a look.

Thanks,
- Tom

Re: [PATCH 1/2] [gdb/tui] Fix segfault in tui_find_disassembly_address

[Tom de Vries]

On 9/28/23 22:57, Tom de Vries wrote:
> On 9/27/23 18:10, Kevin Buettner wrote:
>> Hi Tom,
>>
>> On Tue,  5 Sep 2023 17:03:38 +0200
>> Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> wrote:
>>
>>> Fix the segfault by handling asm_lines.empty () results of 
>>> tui_disassemble in
>>> tui_find_disassembly_address.
>>>
>>> I've written a unit test that exercises this scenario.
>>>
>>> Tested on x86_64-linux.
>>>
>>> PR tui/30823
>>> Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=30823
>>
>> I don't know this area very well, but your explanation and patch look
>> reasonable to me.
>>
>> Reviewed-by: Kevin Buettner <kevinb@redhat.com>
> 
> Thanks for the review.
> 
> I've committed this, but I now see there's a new warning:
> ...
> gdb/tui/tui-disasm.c: In function ‘CORE_ADDR 
> tui_find_disassembly_address(gdbarch*, CORE_ADDR, int)’:
> gdb/tui/tui-disasm.c:293:7: warning: ‘last_addr’ may be used 
> uninitialized in this function [-Wmaybe-uninitialized]
>         if (last_addr < pc)
>         ^~
> ...
> 
> I'll take a look.

I've submitted a fix here ( 
https://sourceware.org/pipermail/gdb-patches/2023-September/202849.html ).

Thanks,
- Tom