Shared object corruption to due 32bit roundup()

[eytanaim]

Describe the bug

Using this tool on /lib64/libgcc_s.so.1 (or many other system libraries) on RedHat 5, resulted corruption to the shared object.

After investigating the issue, I believe I have found the root-cause:
The roundup function, handles int values (32bit) while it is sometimes used with 64bit values:

void ElfFile<ElfFileParamNames>::rewriteSectionsLibrary()
{
    /* For dynamic libraries, we just place the replacement sections
       at the end of the file.  They're mapped into memory by a
       PT_LOAD segment located directly after the last virtual address
       page of other segments. */
    Elf_Addr startPage = 0;
    for (unsigned int i = 0; i < phdrs.size(); ++i) {
        Elf_Addr thisPage = roundUp(rdi(phdrs[i].p_vaddr) + rdi(phdrs[i].p_memsz), getPageSize());
        if (thisPage > startPage) startPage = thisPage;
    }
...

When phdrs[i].p_vaddr requires 64 bit representation, it leads to the corruption.

Steps To Reproduce

• Extract my files myElfs.zip
• Run patchelf --debug --add-needed eytan.so ./libgcc_s.so.1
• File is now corrupted even if it finds eytans.so :)
• Output of following commands fails (needless to say libgcc is a dynamic executable):

[root@agentcos5smp3 wip]# ldd  libgcc_s.so.1
        not a dynamic executable
[root@agentcos5smp3 wip]# readelf -a libgcc_s.so.1 > /dev/null
readelf: Warning: Virtual address 0x377ac68168 not located in any PT_LOAD segment.
readelf: Error: Unable to read in dynamic data

Comparing the readelf output of before and after shows the issue:
The 64bit value of 0x000000377ae***** is being trimmed to 0x000000007ae***** due to a bad roundup

Expected behavior
File to valid

patchelf --version output
patchelf 0.11

Additional context

Add any other context about the problem here.

[DerDakon]

This should help:

--- a/src/patchelf.cc
+++ b/src/patchelf.cc
@@ -541,7 +541,7 @@ static void writeFile(std::string fileName, FileContents contents)
 }
 
 
-static unsigned int roundUp(unsigned int n, unsigned int m)
+static uint64_t roundUp(uint64_t n, uint64_t m)
 {
     return ((n - 1) / m + 1) * m;
 }

[eytanaim]

Any reason why this patch is not applied into master?
When fixing this issue I found 2 additional things. I'll update once I find out the RCA.

[eytanaim]

After using the roundup for 64 bit I found out 2 additional issues that relates to the same issue - the fact that the offset value is not equal to the address value for the section and program headers in the files I have attached above.

The following patch fixes both of them. I'm not sure if there are others issues that might raise using the library above.

patch.txt