The main motivation for this change is to handle very large stack
traces, normally the result of infinite recursion. This part is
actually fairly simple, relaxing a few self-imposed limits on how
many frames we can unwind and the max size for stack memory.
Relaxing these limits requires stricter and more consistent checks for
stack unwinding. There are a number of unwinding invariants that apply
to all the platforms:
1. stack pointer (and frame pointer) must be within the stack memory
(frame pointer, if preset, must point to the right frame too)
2. unwinding must monotonically increase SP
(except for the first frame unwind, this must be a strict increase)
3. Instruction pointer (return address) must point to a valid location
4. stack pointer (and frame pointer) must be appropriately aligned
This change is focused on 2), which is enough to guarantee that the
unwinding doesn't get stuck in an infinite loop.
1) is implicitly validated part of accessing the stack memory
(explicit checks might be nice though).
4) is ABI specific and while it may be valuable in catching suspicious
frames is not in the scope of this change.
3) is also an interesting check but thanks to just-in-time compilation
it's more complex than just calling
StackWalker::InstructionAddressSeemsValid()
and we don't want to drop parts of the callstack due to an overly
conservative check.
Bug: chromium:735989
Change-Id: I9aaba77c7fd028942d77c87d51b5e6f94e136ddd
Reviewed-on: https://chromium-review.googlesource.com/563771
Reviewed-by: Mark Mentovai <mark@chromium.org>
Reviewed-by: Ivan Penkov <ivanpe@chromium.org>
The x86-64 frame pointer-based unwind method will accept values
that aren't valid for the frame pointer register and the return address.
This fixes it to reject non-8-byte-aligned frame pointers, as
well as non-canonical addresses for the return address it finds.
A colleague of mine asked me why Breakpad gave a bad stack
for a crash in our crash-stats system:
https://crash-stats.mozilla.com/report/index/a472c842-2c7b-4ca7-a267-478cf2160405
Digging in, it turns out that the function in frame 0 is a leaf function,
so MSVC doesn't generate an entry in the unwind table for it, so
dump_syms doesn't produce a STACK CFI entry for it in the symbol file.
The stackwalker tries frame pointer unwinding, and %rbp is set to a
value that sort-of works, so it produces a garbage frame 1 and then
is lost. Either of the two checks in this patch would have stopped
the stackwalker from using the frame pointer.
It's possible we could do something smarter on the dump_syms side,
like enumerating all functions and outputing some default STACK CFI rule
for those that don't have unwind info, but that wouldn't fix crashes
from existing builds without re-dumping symbols for them. In any event,
these checks should always pass for valid frame pointer-using functions.
R=mark@chromium.org
BUG=https://bugzilla.mozilla.org/show_bug.cgi?id=1263001
Review URL: https://codereview.chromium.org/1902783002 .
There is an issue in StackwalkerAMD64::GetCallerByFramePointerRecovery.
Occasionally it produces invalid frames (instruction pointer == 0) which
prevents the AMD64 stack walker from proceeding to do stack scanning and
instead leads to premature termination of the stack walking process.
For more details: http://crbug/537444
BUG=
R=mark@chromium.org
Review URL: https://codereview.chromium.org/1408973002 .
This is achieved by:
1. Extending the span of the scan for return address in the conext frame. Initially, I wanted to extend the span of the scan for all frames but then I noticed that there is code for ARM already that is extending the search only for the context frame. This kind of makes sense so I decided to reuse the same idea everywhere.
2. Attempting to restore the EBP chain after a successful scan for return address so that the stackwalker can switch back to FRAME_TRUST_CFI for the rest of the frames when possible.
I also fixed the lint errors in the files touched.
Review URL: https://breakpad.appspot.com/605002
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@1193 4c0a9323-5329-0410-9bdc-e9ce6186880e
This adds support for 'STACK CFI' records (DWARF CFI) to the AMD64
stack walker. This is necessary for the stack trace to include any
frames other than the youngest. Unit tests are included.
a=jimblandy, r=mmentovai
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@554 4c0a9323-5329-0410-9bdc-e9ce6186880e
We've gotten mixed advice from the lawyery types about whether this
matters. But it's easy enough to do.
a=jimblandy, r=nealsid
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@517 4c0a9323-5329-0410-9bdc-e9ce6186880e
At the moment, the StackWalker GetCallerFrame member function expects
a vector of WindowsFrameInfo structures, even though WindowsFrameInfo
is only used or useful on one one implementation (StackWalkerX86).
This patch changes StackWalker::GetCallerFrame to no longer expect the
WindowsFrameInfo structures, and changes all implementations to match.
In particular, StackWalkerX86 is changed to find the WindowsFrameInfo
data itself, and store a pointer to whatever it got in the StackFrame
object itself (which is really a StackFrameX86).
To allow GetCallerFrame implementations to look up stack walking data,
StackWalker::resolver_ needs to be made protected, not private.
a=jimblandy, r=mmentovai
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@491 4c0a9323-5329-0410-9bdc-e9ce6186880e
Also, rename stack_frame_info.h to windows_frame_info.h.
If it seems odd to have functions like FillSourceLineInfo returning
Windows-specific data structures... well, it is! This patch just makes
it more obvious what's going on.
a=jimblandy, r=nealsid
git-svn-id: http://google-breakpad.googlecode.com/svn/trunk@471 4c0a9323-5329-0410-9bdc-e9ce6186880e
