Linking and Loading

1
Linking and Loading

• Fred Prussack
• CS 518

2
LL Overview

• Wake-up Questions
• Terms and Definitions / General Information
• Loading
• Linking
• Static vs. Dynamic
• ELF
• Other Linking Information/Processing
• References

3
LL Topics Not Covered

• Windows
• Alternate binary file formats
• Different versions of glibc
• Different versions of the kernel (from 2.4)

4
LL Questions for the Sleepy

• What is the name of the compile time linker in
  Linux?
• ld
• What is the name of the run-time linker in Linux?
• ld.so
• Where is the loader located in Linux?
• Part kernel / part ld.so

5
LL Terms

• Linking
• Taking object files and creating loadable modules
  with correct references to required libraries,
  data, and procedures
• Loading
• Placing a program image into main memory for
  execution

6
LL General Information

• Static libraries generally named xxx.a (archives)
• Dynamic libraries generally named xxx.so (shared
  objects)
• Object files generally named xxx.o
• ELF Executable and Linking Format. Currently the
  most common object file format on Linux systems.
  Other formats a.out, COFF, etc

7
LL Static vs. Dynamic

• Fully statically compiled executables
• Provide for faster load-gtexecution time due to no
  run-time linking requirement
• Generate larger executables requiring more disk
  space
• Executables with dynamic dependencies
• Require run-time linking and thus potential time
  implications
• Allows for easier and better code re-use

8
LL Loading
do_execve()
Searches all the registered binary handlers
search_binary_handler
Loads current binary and elf interpreter
load_elf_binary
sets up correct registers
start_thread
Question What does the instruction pointer have
in it now?
Answer Entry point of the ELF interpreter
9
LL Loading/Linking

• At this point ld.so now has control
• Determine what libraries need to be loaded for
  this binary
• Determine dependencies for these libraries
• In what order are these loaded and what type of
  list is produced from this dependency list?

10
LL Linking

• Basic job to clean up unresolved symbols
• At compile time this can be accomplished by
  executing ld with object files to produce an
  executable
• At run-time this is accomplished by loading all
  required shared libraries (.sos) and fixing
  unresolved symbols found in the libraries

11
LL Dynamic Linking

• Load Time Dynamic Linking Done By ld.so
• Most likely on your system it is ld-linux.so
  which links to ld-2.3.2.so
• All possibly resolved symbols are resolved during
  compilation/first link (run of ld). Remaining
  unresolved symbols are done at time of load
• Lazy Binding (LD_BIND_NOW)
• Run Time Dynamic (inline) Linking
• Allows applications to, during run time, open
  shared object files and execute their functions
• ltdlfcn.hgt

12
LL more Questions for the Sleepy

• In what package is ld.so distributed and built
  from?
• Glibc
• Can gcc be made to not link files automatically?
• Yes, of course! Use the c option.

13
LL ELF File Format

• Currently the standard binary format for Linux
  since the late 90s. Created in late 80s.
• Three types of object files
• Shared Object Files (.so shared object file)
• Relocatable Object Files (.o object file)
• Executable Object Files (executable binary file)
• First 4 characters of this type of file is
  backspace (ascii 127)ELF

14
LL ELF File Format

• First the ELF Header 52 bytes in length on a 32
  bit system
• Sections and Segments for libraries and binaries
• Various ELF segments
• text program instructions
• data initialized data
• plt procedure linkage table
• got global offset table
• Checking for NEEDED entries in the dynamic
  segment will let ld know what it needs to load

15
LL ld.so Library Location

• ld.so must be able to correctly locate the
  identified libraries in the executable. It does
  this by looking for them in the following order
• DT_RPATH (-rpath-link option)
• Section in ELF file
• LD_LIBRARY_PATH
• Environment Variable
• /etc/ld.so.cache
• Compiled list of files to load
• /lib /usr/lib
• /etc/ld.so.conf

16
LL ld.so processing

• Loop all the program headers to find necessary
  info
• PHDR (program header) where the program headers
  start This must be found first.
• DYNAMIC indicates where to find the dynamic
  segment (what must be loaded)
• NEEDED Name of file needed
• INTERP used to find the interpreter which
  generally turns out to be ld.so

17
LL ld.so processing

• Load all required libraries found in NEEDED
  portions of the DYNAMIC segment
• Get all necessary information from library
• Dynamic header phdr load header

18
LL ld.so info
Read-Only
Read-Write
19
LL ld.so processing

• What about when we actually call a function that
  hasnt been loaded?
• First need to resolve addressing issues
• Probably best to permanently fix them
• Then we need to call the actual procedure

20
LL ld.so processing
Next Procedure Run
PLT0 pushl GOT 4 jmp GOT
8 PLTN jmp GOTn push
reloc_offset jmp PLT0
library
Routine to fix GOT then jump to procedure after
locating correct symbol
reloc_offset
Stack
Procedure Start Loc
Question What is the name of the fix routine?
Answer fixup
21
LL ld.so misc. info

• You can run ld.so from the command line with an
  executable
• This provides a great ability to test out new
  ld.sos if necessary
• /lib/ld-linux.so executable args

22
LL linking helper tools

• ldd list the dynamic dependancies
• readelf displays information from ELF files
• objdump show information from object files
• nm show symbol information from object files
• strip removes symbols from object files
• LD_DEBUG/LD_DEBUG_OUTPUT shows debug output
  from ld.so

23
LL References

• Stallings, William. Operating Systems Internals
  and Design Principles, 4th Edition. Upper Saddle
  River, NJ Prentice-Hall, 2001
• http//efrw01.frascati.enea.it/Software/Unix/IstrF
  TU/cern-cnl-2001-003-25-link.html
• http//www.iecc.com/linker/linker10.html
• http//www.ibiblio.org/oswg/oswg-nightly/oswg/en_G
  B.ISO_8859-1/books/linux-c-programming/GCC-HOWTO/x
  796.html
• http//linux.about.com/library/cmd/blcmdl2_execve.
  htm
• http//www.iecc.com/linker/
• http//www.suse.de/bastian/Export/linking.txt
• http//linux.about.com/library/cmd/blcmdl8_ld.so.h
  tm
• http//www.linuxjournal.com/node/6463
• http//www.ibiblio.org/oswg/oswg-nightly/oswg/en_G
  B.ISO_8859-1/books/linux-c-programming/GCC-HOWTO/x
  575.html

24
LL References (cont.)

• http//www.moses.uklinux.net/patches/lki-single.ht
  ml
• http//whatis.techtarget.com/definition/0,,sid9_gc
  i212493,00.html
• http//encyclopedia.thefreedictionary.com/position
  20independent20code
• http//www.faqs.org/docs/Linux-HOWTO/Program-Libra
  ry-HOWTO.html
• http//sources.redhat.com/autobook/autobook/autobo
  ok_71.html
• http//www.educ.umu.se/bjorn/linux/howto/ELF-HOWT
  O-1.html
• http//www.tcfs.it/docs/manpages/BSD/gcc-howto-6.h
  tml
• http//www.cs.ucdavis.edu/haungs/paper/
• http//www-106.ibm.com/developerworks/linux/librar
  y/l-dll.html?dwzonelinux