[object] Fix crash when input file has no contents#1166
Conversation
| return true; | ||
| if (!Input->getSize()) | ||
| if (!Input->getSize()) { | ||
| ThisConfig.raise(Diag::input_file_has_zero_size) << Input->decoratedPath(); |
There was a problem hiding this comment.
Are we emitting this diagnostic for non-existing file? If so, that is incorrect. Also with this modification, the link will fail for existing but 0-sized files. Is this an intended behavior change?
There was a problem hiding this comment.
Hi @parth-07, I think you are right in both cases. I will update the checks.
| # Verify that ELD does not crash when a linker script references a | ||
| # file that cannot be read (e.g. not found in sysroot). | ||
| # ELD should error out gracefully instead of segfaulting. | ||
| # Fix for issue #818. |
There was a problem hiding this comment.
github specific references can probably be left out.
There was a problem hiding this comment.
Hi @quic-areg, Thank You ::)). Will remove it.
deepakshirkem
force-pushed
the
fix/sysroot-script-crash-818
branch
2 times, most recently
from
ea77759 to
02f9cdf
Compare
|
Hi @parth-07, I have updated the fix. Here is the verified behavior Hi @quic-areg , Updated the test. Please go one more time. |
deepakshirkem
force-pushed
the
fix/sysroot-script-crash-818
branch
from
02f9cdf to
b8fb315
Compare
|
Hi @Parth / @quic-areg / @quic-seaswara , Updated the suggested test. Please review when get a chance and also the run pipeline. |
| RUN: %clang %clangopts -o %t1.o %p/Inputs/1.c -c | ||
| RUN: touch %t1.empty.o | ||
| RUN: llvm-ar rcs %t1.a %t1.o %t1.empty.o | ||
| RUN: llvm-ar rcsT %t1.thin.a %t1.o %t1.empty.o |
There was a problem hiding this comment.
Use the lit variables, do we know if this will work on windows ?
| RUN: %clang %clangopts -o %t1.o %p/Inputs/main.c -c | ||
| RUN: %not %link %linkopts %t1.o -T %p/Inputs/script.t 2>&1 | %filecheck %s | ||
| #END_TEST | ||
| CHECK: cannot read file |
deepakshirkem
force-pushed
the
fix/sysroot-script-crash-818
branch
from
b8fb315 to
fdfcc03
Compare
|
Hi @quic-seaswara, addressed your comments. |
deepakshirkem
force-pushed
the
fix/sysroot-script-crash-818
branch
from
fdfcc03 to
62a09af
Compare
|
Hi @quic-seaswara, Can you re-run the pipeline one more time. |
ELD crashes with a segfault when a linker script references a file that cannot be read (e.g. not found in sysroot during reproduce tarball re-run). The crash occurs because readAndProcessInput() raises the input_file_has_zero_size diagnostic but continues processing the empty file, eventually crashing in ScriptLexer when it tries to dereference a null MemoryArea. This avoids a behavior change for existing 0-sized files. Fixes qualcomm#818. Signed-off-by: deepakshirkem <deepakshirke509@gmail.com>
deepakshirkem
force-pushed
the
fix/sysroot-script-crash-818
branch
from
62a09af to
937fa19
Compare
|
Hi @quic-seaswara, I am not sure why the CI failed for that test case, as it is passing locally for me. I think it might be because my branch was out of date. I have now rebased it with the latest top/main branch. Could you please run the CI one more time? Sorry |
| // 2. File does not exist (MemArea = null): | ||
| // fatal_cannot_read_input already raised. Return false to | ||
| // prevent crash in downstream processing. | ||
| if (Input->getMemArea()) |
There was a problem hiding this comment.
We should not reach readAndProcessInput for an invalid Input *. We resolve the input path and create the MemoryArea for the input in Input::resolvePath(). This function should report the error and return false if the file does not exist.
4 participants
Problem
ELD crashes with a segfault when a reproduce tarball contains a linker script that references a file not found in the sysroot.
Testing
Added test
SysrootScriptCrashthat verifies ELD errors out gracefully instead of segfaulting when a linker script references an unreadable file.Screenshot
Fixes #818
cc @quic-seaswara , @parth-07 , @quic-areg