Hello,
I found a bug when installing a toolchain sdk and the installation
path contains a dash.
Using the default directory or a path without a dash it works.

Yocto 3.0 'zeus'

$ ./fslc-x11-glibc-x86_64-meta-toolchain-cortexa9t2hf-neon-koan-toolchain-3.0.sh
FSLC X11 SDK installer version 3.0
==================================
Enter target directory for SDK (default: /opt/fslc-x11/3.0): /opt/poky/test-3.0
You are about to install the SDK to "/opt/poky/test-3.0". Proceed [Y/n]?
Extracting SDK...............................................................................................................done
Setting it up...ERROR: could not relocate
/opt/poky/test-3.0/sysroots/x86_64-fslcsdk-linux/usr/bin/llvm-mc,
interp size = 73 and 76 is needed.
ERROR: could not relocate
/opt/poky/test-3.0/sysroots/x86_64-fslcsdk-linux/usr/bin/llvm-mca,
interp size = 73 and 76 is needed.
ERROR: could not relocate
/opt/poky/test-3.0/sysroots/x86_64-fslcsdk-linux/usr/bin/clang-9,
interp size = 73 and 76 is needed.
ERROR: could not relocate
/opt/poky/test-3.0/sysroots/x86_64-fslcsdk-linux/usr/bin/llvm-mt,
interp size = 73 and 76 is needed.
...etc...

Distinti Saluti / Best Regards
--
Marco Cavallini | KOAN sas | Bergamo - Italia
embedded software engineering
https://KoanSoftware.com

On Tue, Jan 28, 2020 at 09:25:18AM +0100, Marco Cavallini wrote:

Hello,
I found a bug when installing a toolchain sdk and the installation
path contains a dash.
Using the default directory or a path without a dash it works.

It's not dash, it's the new path is longer than what was reserved in ELF
headers:

interp size = 73 and 76 is needed.

I thought there was some attempt to use chrpath or patchelf or something
to be able to expand the ELF headers with longer paths...

Not sure what the currect status is - I've been using my own workaround
for many years that extends the "default" toolchain location to the max
path length and then at the end changes the install script back to the
original default. That way all ELF headers have "interp" field pre-allocated
to the max and the resulting SDK/toolchain can be relocated w/o issues.

--
Denys

Yocto 3.0 'zeus'

$ ./fslc-x11-glibc-x86_64-meta-toolchain-cortexa9t2hf-neon-koan-toolchain-3.0.sh
FSLC X11 SDK installer version 3.0
==================================
Enter target directory for SDK (default: /opt/fslc-x11/3.0): /opt/poky/test-3.0
You are about to install the SDK to "/opt/poky/test-3.0". Proceed [Y/n]?
Extracting SDK...............................................................................................................done
Setting it up...ERROR: could not relocate
/opt/poky/test-3.0/sysroots/x86_64-fslcsdk-linux/usr/bin/llvm-mc,
interp size = 73 and 76 is needed.
ERROR: could not relocate
/opt/poky/test-3.0/sysroots/x86_64-fslcsdk-linux/usr/bin/llvm-mca,
interp size = 73 and 76 is needed.
ERROR: could not relocate
/opt/poky/test-3.0/sysroots/x86_64-fslcsdk-linux/usr/bin/clang-9,
interp size = 73 and 76 is needed.
ERROR: could not relocate
/opt/poky/test-3.0/sysroots/x86_64-fslcsdk-linux/usr/bin/llvm-mt,
interp size = 73 and 76 is needed.
...etc...

Distinti Saluti / Best Regards
--
Marco Cavallini | KOAN sas | Bergamo - Italia
embedded software engineering
https://KoanSoftware.com

Hello,

I am also facing the same issue with path size. Can you share the workaround what you are using for this issue?

In the scripts/relocate_sdk.py the file name comparison is like below:

if (len(new_dl_path) >= p_filesz):

                print("ERROR: could not relocate %s, interp size = %i and %i is needed." \

                    % (elf_file_name, p_memsz, len(new_dl_path) + 1))

                break

            dl_path = new_dl_path + b("\0") * (p_filesz - len(new_dl_path))

And, to fix the issue I made changes like below.

if (len(new_dl_path) >= 4096):

                print("ERROR: could not relocate %s, interp size = %i and %i is needed." \

                    % (elf_file_name, p_memsz, len(new_dl_path) + 1))

                break

            dl_path = new_dl_path + b("\0") * (4096 - len(new_dl_path))

and commented below code.

#if old_size != os.path.getsize(e):

        #print("New file size for %s is different. Looks like a relocation error!", e)

        #sys.exit(-1)

Do you have any clue regarding,
- Why the installation path is depending of elf headers i.e., Why installation error when the len(new_dl_path) is greater than p_filesz?
- Changing this comparison (len(new_dl_path) >= 4096) will impact the installed SDK?

Hello,

I am also facing the same issue with path size. Can you share the workaround what you are using for this issue?

In the scripts/relocate_sdk.py the file name comparison is like below:

if (len(new_dl_path) >= p_filesz):

                print("ERROR: could not relocate %s, interp size = %i and %i is needed." \

                    % (elf_file_name, p_memsz, len(new_dl_path) + 1))

                break

            dl_path = new_dl_path + b("\0") * (p_filesz - len(new_dl_path))

And, to fix the issue I made changes like below.

if (len(new_dl_path) >= 4096):

                print("ERROR: could not relocate %s, interp size = %i and %i is needed." \

                    % (elf_file_name, p_memsz, len(new_dl_path) + 1))

                break

            dl_path = new_dl_path + b("\0") * (4096 - len(new_dl_path))

and commented below code.

#if old_size != os.path.getsize(e):

        #print("New file size for %s is different. Looks like a relocation error!", e)

        #sys.exit(-1)

Do you have any clue regarding,
- Why the installation path is depending of elf headers i.e., Why installation error when the len(new_dl_path) is greater than p_filesz?
- Changing this comparison (len(new_dl_path) >= 4096) will impact the installed SDK?

The 4096 comes from PATH_MAX:

linux/limits.h.
#define PATH_MAX 4096 /* # chars in a path name including nul */

but that does seem a tad on the long side. I'd prefer a more reasonable
limit that we have actually tested, say PATH_MAX/4 (??) and a comment in the
commit log documenting the choice.

Finally, we do have a test that extracts the sdk in:

   meta/lib/oeqa/sdk/testsdk.py

Can you review that code and use a long but fixed length path by
default so that we catch regressions? The path is set in:
        sdk_dir = d.expand("${WORKDIR}/testimage-sdk/")
since len(${WORKDIR} may vary, I'd like to see code that corrects
for that and results in a fix length of PATH_MAX/4 assuming that
things work with such a long path.


Denis? Does that make sense and what is your work-around?


../Randy


 

-- 
# Randy MacLeod
# Wind River Linux