This is an archive of the discontinued LLVM Phabricator instance.

Differential D124950

[HWASan] Allow to linkify symbolizer output.
ClosedPublic

Authored by fmayer on May 4 2022, 11:27 AM.

Details

Reviewers
eugenis
Commits
rGd8564dcbcf72: [HWASan] Allow to linkify symbolizer output.
rG4af9392e13a2: [HWASan] Allow to linkify symbolizer output.

Diff Detail

Event Timeline

fmayer created this revision.May 4 2022, 11:27 AM
Herald added a project: Restricted Project. · View Herald TranscriptMay 4 2022, 11:27 AM
fmayer updated this revision to Diff 427093.May 4 2022, 11:37 AM

style

fmayer updated this revision to Diff 427095.May 4 2022, 11:38 AM

fix

Harbormaster completed remote builds in B162764: Diff 427095.May 4 2022, 12:27 PM
fmayer published this revision for review.May 4 2022, 2:39 PM
fmayer added a reviewer: eugenis.
Herald added a project: Restricted Project. · View Herald TranscriptMay 4 2022, 2:40 PM
Herald added a subscriber: Restricted Project. · View Herald Transcript
eugenis added a comment.May 4 2022, 3:53 PM

This script is starting to get complex. Perhaps we should write tests?

fmayer added a comment.May 4 2022, 3:54 PM
In D124950#3492520, @eugenis wrote:

This script is starting to get complex. Perhaps we should write tests?

Yes, that is on my TODO list. Happy to do that before submission, but wanted to make sure this approach is fine before doing so.

eugenis added a comment.May 4 2022, 3:56 PM

The approach looks fine. Should we also put a <html></html> pair around the output, or do you want to treat it as a snippet?

fmayer added a comment.May 4 2022, 4:00 PM
In D124950#3492544, @eugenis wrote:

The approach looks fine. Should we also put a <html></html> pair around the output, or do you want to treat it as a snippet?

Generally I don't think this would be used a standalone input, so maybe a snippet is better and the consumer can add whatever they want around. For my use-case this gets piped into pandoc to be converted to Markdown.

fmayer updated this revision to Diff 427193.May 4 2022, 9:42 PM

Add test

fmayer added a comment.May 4 2022, 9:43 PM
In D124950#3492520, @eugenis wrote:

This script is starting to get complex. Perhaps we should write tests?

done. added a test for general symbolization and for linkifying

Harbormaster completed remote builds in B162828: Diff 427193.May 4 2022, 10:01 PM
fmayer added a child revision: D125032: [HWASan] Clean up hwasan_symbolize..May 5 2022, 11:54 AM
eugenis added inline comments.May 5 2022, 2:46 PM
compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
3

trying to think if this will run correctly on android, when the test binary runs on the device, but FileCheck runs on the host. Also, I think we push llvm-symbolizer to the device for online symbolization, you may need to disable it through HWASAN_OPTIONS.

fmayer added inline comments.May 5 2022, 2:48 PM
compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
3

It does. What happens is that %t becomes the wrapper script (and the real binary becomes %t.real) that runs on the device. The whole pipeline specified here happens on the host.

I ran this on Android.

fmayer added inline comments.May 5 2022, 2:49 PM
compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
3

Good point on the HWASAN_OPTIONS. That makes it more resilient.

eugenis accepted this revision.May 5 2022, 2:55 PM

LGTM

This revision is now accepted and ready to land.May 5 2022, 2:55 PM
fmayer updated this revision to Diff 427466.May 5 2022, 2:56 PM

address comment

Harbormaster completed remote builds in B163022: Diff 427466.May 5 2022, 3:26 PM
Closed by commit rG4af9392e13a2: [HWASan] Allow to linkify symbolizer output. (authored by fmayer). · Explain WhyMay 5 2022, 4:40 PM
This revision was automatically updated to reflect the committed changes.
fmayer added a commit: rG4af9392e13a2: [HWASan] Allow to linkify symbolizer output..
dyung added a subscriber: dyung.May 6 2022, 12:41 AM

Hi, the test you added, hwasan_symbolize.cpp seems to be causing a failure on the sanitizer buildbot (https://lab.llvm.org/buildbot/#/builders/37/builds/13022):

FAIL: HWAddressSanitizer-x86_64 :: TestCases/hwasan_symbolize.cpp (71 of 86)
******************** TEST 'HWAddressSanitizer-x86_64 :: TestCases/hwasan_symbolize.cpp' FAILED ********************
Script:
--
: 'RUN: at line 1';      /b/sanitizer-x86_64-linux/build/clang_build/./bin/clang   -m64  -gline-tables-only -fsanitize=hwaddress -fuse-ld=lld -fsanitize-hwaddress-experimental-aliasing -mllvm -hwasan-generate-tags-with-calls=1 -mllvm -hwasan-globals -mllvm -hwasan-use-short-granules -mllvm -hwasan-instrument-landing-pads=0 -mllvm -hwasan-instrument-personality-functions -Wl,--build-id -g /b/sanitizer-x86_64-linux/build/llvm-project/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp -o /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp
: 'RUN: at line 2';   echo '[{"prefix": "'"$(realpath $(dirname /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp)/../../../../../../)"'/", "link": "http://test.invalid/{file}:{line}"}]' > /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp.linkify
: 'RUN: at line 3';   env HWASAN_OPTIONS=disable_allocator_tagging=1:random_tags=0:fail_without_syscall_abi=0:symbolize=0 not  /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp 2>&1 | hwasan_symbolize --html --symbols $(dirname /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp) --index | FileCheck /b/sanitizer-x86_64-linux/build/llvm-project/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
: 'RUN: at line 4';   env HWASAN_OPTIONS=disable_allocator_tagging=1:random_tags=0:fail_without_syscall_abi=0:symbolize=0 not  /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp 2>&1 | hwasan_symbolize --html --linkify /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp.linkify --symbols $(dirname /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp) --index | FileCheck --check-prefixes=CHECK,LINKIFY /b/sanitizer-x86_64-linux/build/llvm-project/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
: 'RUN: at line 5';   env HWASAN_OPTIONS=disable_allocator_tagging=1:random_tags=0:fail_without_syscall_abi=0:symbolize=0 not  /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp 2>&1 | hwasan_symbolize --symbols $(dirname /b/sanitizer-x86_64-linux/build/clang_build/projects/compiler-rt/test/hwasan/X86_64/TestCases/Output/hwasan_symbolize.cpp.tmp) --index | FileCheck /b/sanitizer-x86_64-linux/build/llvm-project/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
--
Exit Code: 2
Command Output (stderr):
--
Traceback (most recent call last):
  File "/b/sanitizer-x86_64-linux/build/clang_build/./bin/hwasan_symbolize", line 19, in <module>
    import html
ImportError: No module named html
FileCheck error: '<stdin>' is empty.
FileCheck command line:  FileCheck /b/sanitizer-x86_64-linux/build/llvm-project/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
--

And also our internal linux buildbot is reporting a failure after this commit:

/home/siadmin/jenkins/w/opensource/opensource_build/working_dir/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp:17:14: error: LINKIFY: expected string not found in input
 // LINKIFY: <a href="http://test.invalid/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp:[[@LINE-1]]">
             ^
<stdin>:1:1: note: scanning from here
==27214==ERROR: HWAddressSanitizer: tag-mismatch on address 0x680400000064 at pc 0x55d1c8f6e76b<br/>
^
<stdin>:1:1: note: with "@LINE-1" equal to "16"
==27214==ERROR: HWAddressSanitizer: tag-mismatch on address 0x680400000064 at pc 0x55d1c8f6e76b<br/>
^
<stdin>:6:68: note: possible intended match here
&nbsp;&nbsp;&nbsp;&nbsp;#1 main in /home/siadmin/jenkins/w/opensource/opensource_build/working_dir/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp:16:10<br/>
                                                                   ^

Input file: <stdin>
Check file: /home/siadmin/jenkins/w/opensource/opensource_build/working_dir/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp

-dump-input=help explains the following input dump.

Input was:
<<<<<<
            1: ==27214==ERROR: HWAddressSanitizer: tag-mismatch on address 0x680400000064 at pc 0x55d1c8f6e76b<br/> 
check:17'0     X~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ error: no match found
check:17'1                                                                                                           with "@LINE-1" equal to "16"
            2: READ of size 1 at 0x680400000064 tags: 03/00 (ptr/mem) in thread T0<br/> 
check:17'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            3: &nbsp;&nbsp;&nbsp;&nbsp;#0 SigTrap&lt;0&gt; in [hwasan_rtl]:0<br/> 
check:17'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            4: &nbsp;&nbsp;&nbsp;&nbsp;-&gt; CheckAddress&lt;(__hwasan::ErrorAction)0, (__hwasan::AccessType)0, 0&gt; in [hwasan_rtl]:0<br/> 
check:17'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            5: &nbsp;&nbsp;&nbsp;&nbsp;-&gt; __hwasan_load1 in /home/siadmin/jenkins/w/opensource/opensource_build/working_dir/compiler-rt/lib/hwasan/hwasan.cpp:445:0<br/> 
check:17'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            6: &nbsp;&nbsp;&nbsp;&nbsp;#1 main in /home/siadmin/jenkins/w/opensource/opensource_build/working_dir/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp:16:10<br/> 
check:17'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
check:17'2                                                                        ?                                                                                                  possible intended match
            7: &nbsp;&nbsp;&nbsp;&nbsp;#2 0x7fbc32907bf6 (/lib/x86_64-linux-gnu/libc.so.6+0x21bf6) (BuildId: ce450eb01a5e5acc7ce7b8c2633b02cc1093339e)<br/> 
check:17'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            8: &nbsp;&nbsp;&nbsp;&nbsp;#3 _start in ??:0:0<br/> 
check:17'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            9: <br/> 
check:17'0     ~~~~~~
           10: [0x680400000060,0x680400000080) is a small unallocated heap chunk; size: 32 offset: 4<br/> 
check:17'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           11: <br/> 
check:17'0     ~~~~~~
            .
            .
            .
>>>>>>

Can you please take a look and revert if it will take some time to fix?

thakis added a subscriber: thakis.May 6 2022, 6:51 AM

And also our internal linux buildbot is reporting a failure after this commit:

*snip*

I'm seeing this same failure when I run ninja check-hwasan in my local, fairly vanilla, cmake build.

I'm also seeing it on http://45.33.8.238/linux/75461/step_10.txt . Given the other report of breakage, I'll revert this for now until things are sorted out.

compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
5

Do you know why this works? Normally (in clang and lld and llvm), you have to call llvm_config.add_tool_substitutions() for the binaries that you want lit to find in the build dir. compiler-rt doesn't seem to do this, and hwasan_symbolize seems to be found fine so it works somehow, but I don't know how. Could you explain it to me?

thakis added a reverting change: rGc9faea04b1f8: Revert "[HWASan] Allow to linkify symbolizer output.".May 6 2022, 6:54 AM
thakis added a comment.May 6 2022, 7:19 AM

Here's another bot this failed on, also doing a fairly vanilla cmake build: https://logs.chromium.org/logs/chromium/buildbucket/cr-buildbucket/8814940711487302321/+/u/package_clang/stdout?format=raw

fmayer added a comment.EditedMay 6 2022, 8:31 AM
In D124950#3496660, @thakis wrote:

And also our internal linux buildbot is reporting a failure after this commit:

*snip*

I'm seeing this same failure when I run ninja check-hwasan in my local, fairly vanilla, cmake build.

I'm also seeing it on http://45.33.8.238/linux/75461/step_10.txt . Given the other report of breakage, I'll revert this for now until things are sorted out.

Which version of Python are you running? html is apparently around since 3.2.

EDIT: oh, I misunderstood. "this error" being the FileCheck, not the import. Correct?

thakis added a comment.May 6 2022, 8:34 AM

I'm running python 3.9.10.

As far as I can tell, there are 2 distinct errors:

  1. the import html error on https://lab.llvm.org/buildbot/#/builders/37/builds/13022
  2. FileCheck error seen on @dyung's internal bot, on my machine, and on my bot
fmayer added a comment.May 6 2022, 8:36 AM
In D124950#3496884, @thakis wrote:

I'm running python 3.9.10.

As far as I can tell, there are 2 distinct errors:

  1. the import html error on https://lab.llvm.org/buildbot/#/builders/37/builds/13022
  2. FileCheck error seen on @dyung's internal bot, on my machine, and on my bot

Thanks. Sorry about that, thanks for reverting.

fmayer reopened this revision.May 6 2022, 11:14 AM
This revision is now accepted and ready to land.May 6 2022, 11:14 AM
fmayer planned changes to this revision.May 6 2022, 11:14 AM
fmayer updated this revision to Diff 427768.May 6 2022, 3:13 PM
fmayer marked 2 inline comments as done.

fix test.

This revision is now accepted and ready to land.May 6 2022, 3:13 PM
fmayer added inline comments.May 6 2022, 3:16 PM
compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
5

When I run env in the test, build/bin is the first entry in path, so it makes sense that this works.

Harbormaster completed remote builds in B163242: Diff 427768.May 6 2022, 3:38 PM
fmayer added inline comments.May 6 2022, 3:43 PM
compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp
5

That seems to be due to this in our lit cfg

# Tweak PATH to include llvm tools dir.
if (not config.llvm_tools_dir) or (not os.path.exists(config.llvm_tools_dir)):
  lit_config.fatal("Invalid llvm_tools_dir config attribute: %r" % config.llvm_tools_dir)
path = os.path.pathsep.join((config.llvm_tools_dir, config.environment['PATH']))
config.environment['PATH'] = path
Closed by commit rGd8564dcbcf72: [HWASan] Allow to linkify symbolizer output. (authored by fmayer). · Explain WhyMay 6 2022, 3:44 PM
This revision was automatically updated to reflect the committed changes.
fmayer added a commit: rGd8564dcbcf72: [HWASan] Allow to linkify symbolizer output..

Revision Contents

PathSize
compiler-rt/
lib/
hwasan/
scripts/
85 lines
test/
hwasan/
TestCases/
24 lines

Diff 427491

compiler-rt/lib/hwasan/scripts/hwasan_symbolize

Show All 10 Lines
# #
#===------------------------------------------------------------------------===# #===------------------------------------------------------------------------===#
from __future__ import print_function from __future__ import print_function
from __future__ import unicode_literals from __future__ import unicode_literals
import argparse import argparse
import glob import glob
import html
import json
import mmap import mmap
import os import os
import re import re
import struct import struct
import subprocess import subprocess
import sys import sys
if sys.version_info.major < 3: if sys.version_info.major < 3:
▲ Show 20 Lines • Show All 74 Lines • ▼ Show 20 Linesclass Symbolizer:
def __init__(self, path, binary_prefixes, paths_to_cut): def __init__(self, path, binary_prefixes, paths_to_cut):
self.__pipe = None self.__pipe = None
self.__path = path self.__path = path
self.__binary_prefixes = binary_prefixes self.__binary_prefixes = binary_prefixes
self.__paths_to_cut = paths_to_cut self.__paths_to_cut = paths_to_cut
self.__log = False self.__log = False
self.__warnings = set() self.__warnings = set()
self.__index = {} self.__index = {}
self.__link_prefixes = []
self.__html = False
def enable_html(self, enable):
self.__html = enable
def enable_logging(self, enable): def enable_logging(self, enable):
self.__log = enable self.__log = enable
def maybe_escape(self, text):
if self.__html:
# We need to manually use &nbsp; for leading spaces, html.escape does
# not do that, and HTML ignores them.
spaces = 0
for i, c in enumerate(text):
spaces = i
if c != ' ':
break
text = text[spaces:]
return spaces * '&nbsp;' + html.escape(text)
return text
def print(self, line, escape=True):
if escape:
line = self.maybe_escape(line)
if self.__html:
line += '<br/>'
print(line)
def read_linkify(self, filename):
with open(filename, 'r') as fd:
data = json.load(fd)
self.__link_prefixes = [(e["prefix"], e["link"]) for e in data]
def __open_pipe(self): def __open_pipe(self):
if not self.__pipe: if not self.__pipe:
opt = {} opt = {}
if sys.version_info.major > 2: if sys.version_info.major > 2:
opt['encoding'] = 'utf-8' opt['encoding'] = 'utf-8'
self.__pipe = subprocess.Popen([self.__path, "--inlining", "--functions"], self.__pipe = subprocess.Popen([self.__path, "--inlining", "--functions"],
stdin=subprocess.PIPE, stdout=subprocess.PIPE, stdin=subprocess.PIPE, stdout=subprocess.PIPE,
**opt) **opt)
▲ Show 20 Lines • Show All 81 Lines • ▼ Show 20 Lines try:
while True: while True:
function_name = self.__read() function_name = self.__read()
file_line = self.__read() file_line = self.__read()
file_line = self.__process_source_path(file_line) file_line = self.__process_source_path(file_line)
yield (function_name, file_line) yield (function_name, file_line)
except Symbolizer.__EOF: except Symbolizer.__EOF:
pass pass
def maybe_linkify(self, file_line):
if not self.__html or not self.__link_prefixes:
return file_line
filename, line_col = file_line.split(':', 1)
if not line_col:
line = '0' # simplify the link generation
else:
line = line_col.split(':')[0]
longest_prefix = max((
(prefix, link) for prefix, link in self.__link_prefixes
if filename.startswith(prefix)),
key=lambda x: len(x[0]), default=None)
if longest_prefix is None:
return file_line
else:
prefix, link = longest_prefix
return '<a href="{}">{}</a>'.format(
html.escape(link.format(file=filename[len(prefix):], line=line,
file_line=file_line, prefix=prefix)), file_line)
def build_index(self): def build_index(self):
for p in self.__binary_prefixes: for p in self.__binary_prefixes:
for dname, _, fnames in os.walk(p): for dname, _, fnames in os.walk(p):
for fn in fnames: for fn in fnames:
filename = os.path.join(dname, fn) filename = os.path.join(dname, fn)
bid = get_buildid(filename) bid = get_buildid(filename)
if bid is not None: if bid is not None:
self.__index[bid] = filename self.__index[bid] = filename
def symbolize_line(line, symbolizer_path): def symbolize_line(line, symbolizer_path):
#0 0x7f6e35cf2e45 (/blah/foo.so+0x11fe45) (BuildId: 4abce4cd41ea5c2f34753297b7e774d9) #0 0x7f6e35cf2e45 (/blah/foo.so+0x11fe45) (BuildId: 4abce4cd41ea5c2f34753297b7e774d9)
match = re.match(r'^(.*?)#([0-9]+)( *)(0x[0-9a-f]*) *\((.*)\+(0x[0-9a-f]+)\)' match = re.match(r'^(.*?)#([0-9]+)( *)(0x[0-9a-f]*) *\((.*)\+(0x[0-9a-f]+)\)'
r'(?:\s*\(BuildId: ([0-9a-f]+)\))?', line, re.UNICODE) r'(?:\s*\(BuildId: ([0-9a-f]+)\))?', line, re.UNICODE)
if match: if match:
frameno = match.group(2) frameno = match.group(2)
binary = match.group(5) binary = match.group(5)
addr = int(match.group(6), 16) addr = int(match.group(6), 16)
buildid = match.group(7) buildid = match.group(7)
frames = list(symbolizer.iter_call_stack(binary, buildid, addr)) frames = list(symbolizer.iter_call_stack(binary, buildid, addr))
if len(frames) > 0: if len(frames) > 0:
print("%s#%s%s%s in %s" % (match.group(1), match.group(2), symbolizer.print(
match.group(3), frames[0][0], frames[0][1])) symbolizer.maybe_escape(
"%s#%s%s%s in " % (match.group(1), match.group(2), match.group(3),
frames[0][0])
) + symbolizer.maybe_linkify(frames[0][1]),
escape=False)
for i in range(1, len(frames)): for i in range(1, len(frames)):
space1 = ' ' * match.end(1) space1 = ' ' * match.end(1)
space2 = ' ' * (match.start(4) - match.end(1) - 2) space2 = ' ' * (match.start(4) - match.end(1) - 2)
print("%s->%s%s in %s" % (space1, space2, frames[i][0], frames[i][1])) symbolizer.print(
symbolizer.maybe_escape("%s->%s%s in " % (space1, space2, frames[i][0]))
+ symbolizer.maybe_linkify(frames[i][1]), escape=False)
else: else:
print(line.rstrip()) symbolizer.print(line.rstrip())
else: else:
print(line.rstrip()) symbolizer.print(line.rstrip())
def save_access_address(line): def save_access_address(line):
global last_access_address, last_access_tag global last_access_address, last_access_tag
match = re.match(r'^(.*?)HWAddressSanitizer: tag-mismatch on address (0x[0-9a-f]+) ', line, re.UNICODE) match = re.match(r'^(.*?)HWAddressSanitizer: tag-mismatch on address (0x[0-9a-f]+) ', line, re.UNICODE)
if match: if match:
last_access_address = int(match.group(2), 16) last_access_address = int(match.group(2), 16)
match = re.match(r'^(.*?) of size [0-9]+ at 0x[0-9a-f]* tags: ([0-9a-f]+)/[0-9a-f]+ \(ptr/mem\)', line, re.UNICODE) match = re.match(r'^(.*?) of size [0-9]+ at 0x[0-9a-f]* tags: ([0-9a-f]+)/[0-9a-f]+ \(ptr/mem\)', line, re.UNICODE)
if match: if match:
Show All 25 Lines for local in symbolizer.iter_locals(binary, addr, buildid):
if frame_offset is None or size is None: if frame_offset is None or size is None:
continue continue
obj_offset = (last_access_address - fp - frame_offset) & fp_mask obj_offset = (last_access_address - fp - frame_offset) & fp_mask
if obj_offset >= size: if obj_offset >= size:
continue continue
tag_offset = local[5] tag_offset = local[5]
if not ignore_tags and (tag_offset is None or base_tag ^ tag_offset != last_access_tag): if not ignore_tags and (tag_offset is None or base_tag ^ tag_offset != last_access_tag):
continue continue
print('') symbolizer.print('')
print('Potentially referenced stack object:') symbolizer.print('Potentially referenced stack object:')
print(' %d bytes inside variable "%s" in stack frame of function "%s"' % (obj_offset, local[2], local[0])) symbolizer.print(' %d bytes inside variable "%s" in stack frame of function "%s"' % (obj_offset, local[2], local[0]))
print(' at %s' % (local[1],)) symbolizer.print(' at %s' % (local[1],))
return True return True
return False return False
parser = argparse.ArgumentParser() parser = argparse.ArgumentParser()
parser.add_argument('-d', action='store_true') parser.add_argument('-d', action='store_true')
parser.add_argument('-v', action='store_true') parser.add_argument('-v', action='store_true')
parser.add_argument('--ignore-tags', action='store_true') parser.add_argument('--ignore-tags', action='store_true')
parser.add_argument('--symbols', action='append') parser.add_argument('--symbols', action='append')
parser.add_argument('--source', action='append') parser.add_argument('--source', action='append')
parser.add_argument('--index', action='store_true') parser.add_argument('--index', action='store_true')
parser.add_argument('--symbolizer') parser.add_argument('--symbolizer')
parser.add_argument('--linkify', type=str)
parser.add_argument('--html', action='store_true')
parser.add_argument('args', nargs=argparse.REMAINDER) parser.add_argument('args', nargs=argparse.REMAINDER)
args = parser.parse_args() args = parser.parse_args()
# Unstripped binaries location. # Unstripped binaries location.
binary_prefixes = args.symbols or [] binary_prefixes = args.symbols or []
if not binary_prefixes: if not binary_prefixes:
if 'ANDROID_PRODUCT_OUT' in os.environ: if 'ANDROID_PRODUCT_OUT' in os.environ:
product_out = os.path.join(os.environ['ANDROID_PRODUCT_OUT'], 'symbols') product_out = os.path.join(os.environ['ANDROID_PRODUCT_OUT'], 'symbols')
▲ Show 20 Lines • Show All 69 Lines • ▼ Show 20 Lines for s in binary_prefixes:
print(" %s" % (s,)) print(" %s" % (s,))
print("Stripping source path prefixes:") print("Stripping source path prefixes:")
for s in paths_to_cut: for s in paths_to_cut:
print(" %s" % (s,)) print(" %s" % (s,))
print("Using llvm-symbolizer binary in:\n %s" % (symbolizer_path,)) print("Using llvm-symbolizer binary in:\n %s" % (symbolizer_path,))
print() print()
symbolizer = Symbolizer(symbolizer_path, binary_prefixes, paths_to_cut) symbolizer = Symbolizer(symbolizer_path, binary_prefixes, paths_to_cut)
symbolizer.enable_html(args.html)
symbolizer.enable_logging(args.d) symbolizer.enable_logging(args.d)
if args.index: if args.index:
symbolizer.build_index() symbolizer.build_index()
if args.linkify:
if not args.html:
print('Need --html to --linkify', file=sys.stderr)
sys.exit(1)
symbolizer.read_linkify(args.linkify)
for line in sys.stdin: for line in sys.stdin:
if sys.version_info.major < 3: if sys.version_info.major < 3:
line = line.decode('utf-8') line = line.decode('utf-8')
save_access_address(line) save_access_address(line)
if process_stack_history(line, symbolizer, ignore_tags=args.ignore_tags): if process_stack_history(line, symbolizer, ignore_tags=args.ignore_tags):
continue continue
symbolize_line(line, symbolizer_path) symbolize_line(line, symbolizer_path)

compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp

  • This file was added.
// RUN: %clang_hwasan -Wl,--build-id -g %s -o %t
// RUN: echo '[{"prefix": "'"$(realpath $(dirname %t)/../../../../../../)"'/", "link": "http://test.invalid/{file}:{line}"}]' > %t.linkify
// RUN: %env_hwasan_opts=symbolize=0 not %run %t 2>&1 | hwasan_symbolize --html --symbols $(dirname %t) --index | FileCheck %s
eugenisUnsubmitted
Done
ReplyInline Actions

trying to think if this will run correctly on android, when the test binary runs on the device, but FileCheck runs on the host. Also, I think we push llvm-symbolizer to the device for online symbolization, you may need to disable it through HWASAN_OPTIONS.

eugenis: trying to think if this will run correctly on android, when the test binary runs on the device…
fmayerAuthorUnsubmitted
Done
ReplyInline Actions

It does. What happens is that %t becomes the wrapper script (and the real binary becomes %t.real) that runs on the device. The whole pipeline specified here happens on the host.

I ran this on Android.

fmayer: It does. What happens is that %t becomes the wrapper script (and the real binary becomes %t.
fmayerAuthorUnsubmitted
Done
ReplyInline Actions

Good point on the HWASAN_OPTIONS. That makes it more resilient.

fmayer: Good point on the HWASAN_OPTIONS. That makes it more resilient.
// RUN: %env_hwasan_opts=symbolize=0 not %run %t 2>&1 | hwasan_symbolize --html --linkify %t.linkify --symbols $(dirname %t) --index | FileCheck --check-prefixes=CHECK,LINKIFY %s
// RUN: %env_hwasan_opts=symbolize=0 not %run %t 2>&1 | hwasan_symbolize --symbols $(dirname %t) --index | FileCheck %s
thakisUnsubmitted
Done
ReplyInline Actions

Do you know why this works? Normally (in clang and lld and llvm), you have to call llvm_config.add_tool_substitutions() for the binaries that you want lit to find in the build dir. compiler-rt doesn't seem to do this, and hwasan_symbolize seems to be found fine so it works somehow, but I don't know how. Could you explain it to me?

thakis: Do you know why this works? Normally (in clang and lld and llvm), you have to call `llvm_config.
fmayerAuthorUnsubmitted
Done
ReplyInline Actions

When I run env in the test, build/bin is the first entry in path, so it makes sense that this works.

fmayer: When I run `env` in the test, `build/bin` is the first entry in path, so it makes sense that…
fmayerAuthorUnsubmitted
Done
ReplyInline Actions

That seems to be due to this in our lit cfg

# Tweak PATH to include llvm tools dir.
if (not config.llvm_tools_dir) or (not os.path.exists(config.llvm_tools_dir)):
  lit_config.fatal("Invalid llvm_tools_dir config attribute: %r" % config.llvm_tools_dir)
path = os.path.pathsep.join((config.llvm_tools_dir, config.environment['PATH']))
config.environment['PATH'] = path
fmayer: That seems to be due to this in our lit cfg ``` # Tweak PATH to include llvm tools dir. if…
// REQUIRES: stable-runtime
#include <sanitizer/hwasan_interface.h>
#include <stdlib.h>
static volatile char sink;
int main(int argc, char **argv) {
__hwasan_enable_allocator_tagging();
char *volatile x = (char *)malloc(10);
sink = x[100];
// LINKIFY: <a href="http://test.invalid/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp:[[@LINE-1]]">
// CHECK: hwasan_symbolize.cpp:[[@LINE-2]]
// CHECK: Cause: heap-buffer-overflow
// CHECK: allocated here:
// LINKIFY: <a href="http://test.invalid/compiler-rt/test/hwasan/TestCases/hwasan_symbolize.cpp:[[@LINE-6]]">
// CHECK: hwasan_symbolize.cpp:[[@LINE-7]]
return 0;
}